context("critVal") source(system.file("tests/comparisons/penalisation.R", package = "stepR")) testComputeLevel <- function(q, stat) { M <- ncol(stat) d <- nrow(stat) if (length(q) != d) { stop("dimensions from q and stat are not the same") } rejected <- matrix(FALSE, M, d) alpha <- integer(M) for (i in 1:d) { statVec <- stat[i, ] tol <- 2 * .Machine$double.eps * max(statVec[is.finite(statVec)]) rejected[, i] <- statVec > q[i] + tol } for (i in 1:M) { alpha[i] <- max(rejected[i, ]) } alpha <- sum(alpha) ret <- list(alpha = alpha) beta <- integer(d) for (i in 1:d) { beta[i] <- sum(rejected[, i]) } ret$beta <- beta ret } testDetailedHSMUCE <- function(testq = NULL, testalpha = NULL, testn, testnq = 2L^(as.integer(ceiling(log2(testn)) + 1e-12)), testweights = NULL, testfamily = NULL, testintervalsystem = NULL, testlengths = NULL, teststat = NULL, testr = 1e4, defaultPenalty, compareStat = teststat, compareLengths = testlengths, compareWeights = testweights, tolerance = 1e-12, toleranceInt = 0.5, nstat = testn, shift = 0, ...) { # sqrt ret <- critVal(penalty = "sqrt", output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...) rejected <- matrix(FALSE, ncol(teststat), nrow(teststat)) compare <- integer(ncol(teststat)) for (i in 1:nrow(teststat)) { statVec <- teststat[i, ] tol <- 2 * .Machine$double.eps * max(statVec[is.finite(statVec)]) rejected[, i] <- statVec > ret[i] + tol } for (i in 1:100) { for (j in 2:10) { expect_false(rejected[i, j], info = paste0("i = ", i, ", j = ", j, ": ", teststat[j, i], ", ", ret[j])) } } for (i in (1:100)[-c(2, 21, 24, 30, 34)]) { expect_false(rejected[i, 1], info = paste0("i = ", i, ", j = ", 1, ": ", teststat[1, i], ", ", ret[1])) } for (i in c(2, 21, 24, 30, 34)) { expect_true(rejected[i, 1], info = paste0("i = ", i, ", j = ", 1, ": ", teststat[1, i], ", ", ret[1])) } for (i in 1:ncol(teststat)) { compare[i] <- max(rejected[i, ]) } compare <- sum(compare) expect_lte(compare, as.integer(ncol(teststat) * testalpha + tolerance), label = compare, expected.label = as.integer(ncol(teststat) * testalpha + tolerance)) } testVector <- function(testq = NULL, testalpha = NULL, testn, testnq = 2L^(as.integer(ceiling(log2(testn)) + 1e-12)), testweights = NULL, testfamily = NULL, testintervalsystem = NULL, testlengths = NULL, teststat = NULL, testr = 1e4, defaultPenalty, compareStat = teststat, compareLengths = testlengths, compareWeights = testweights, tolerance = 1e-12, toleranceInt = 0.5, nstat = testn, shift = 0, ...) { # sqrt ret <- critVal(penalty = "sqrt", output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...) expect_true(is.integer(attr(ret, "n"))) expect_identical(attr(ret, "n"), nstat) compare <- testComputeLevel(q = ret, stat = compareStat)$alpha expect_true(compare <= as.integer(ncol(compareStat) * testalpha + tolerance), info = "sqrt") expect_true(compare > as.integer(ncol(compareStat) * testalpha + tolerance) - 1L, info = "sqrt") for (i in 1:length(ret)) { ret[i] <- penalisationSqrt(ret[i], nstat, compareLengths[i] + shift) } expect_equal(ret, rep(ret[1], length(ret)), check.attributes = FALSE, tolerance = tolerance, info = "sqrt") # log ret <- critVal(penalty = "log", output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...) expect_true(is.integer(attr(ret, "n"))) expect_identical(attr(ret, "n"), nstat) compare <- testComputeLevel(q = ret, stat = compareStat)$alpha expect_true(compare <= as.integer(ncol(compareStat) * testalpha + tolerance), info = "log") expect_true(compare > as.integer(ncol(compareStat) * testalpha + tolerance) - 1L, info = "log") for (i in 1:length(ret)) { ret[i] <- penalisationLog(ret[i], nstat, compareLengths[i] + shift) } expect_equal(ret, rep(ret[1], length(ret)), check.attributes = FALSE, tolerance = tolerance, info = "log") # none ret <- critVal(penalty = "none", output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...) expect_true(is.integer(attr(ret, "n"))) expect_identical(attr(ret, "n"), nstat) compare <- testComputeLevel(q = ret, stat = compareStat)$alpha expect_true(compare <= as.integer(ncol(compareStat) * testalpha + tolerance), info = "none") expect_true(compare > as.integer(ncol(compareStat) * testalpha + tolerance) - 1L, info = "none") expect_equal(ret, rep(ret[1], length(ret)), check.attributes = FALSE, tolerance = tolerance, info = "none") # weights ret <- critVal(penalty = "weights", weights = testweights, output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...) expect_true(is.integer(attr(ret, "n"))) expect_identical(attr(ret, "n"), nstat) compare <- testComputeLevel(q = ret, stat = compareStat) expect_true(compare$alpha <= as.integer(ncol(compareStat) * testalpha + tolerance), info = "weights") expect_true(compare$alpha > as.integer(ncol(compareStat) * testalpha + tolerance) - 1L, info = "weights") expect_true(max(compare$beta / compareWeights) - min((compare$beta + 1) / compareWeights) <= 0.5, info = "weights") # default expect_identical(critVal(penalty = defaultPenalty, output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...), critVal(output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...)) } testValue <- function(testq = NULL, testalpha = NULL, testn, testnq = testn, testweights = NULL, testfamily = NULL, testintervalsystem = NULL, testlengths = NULL, teststat = NULL, testr = 1e4, defaultPenalty, tolerance = 1e-12, ...) { # sqrt ret <- critVal(penalty = "sqrt", output = "value", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...) expect_identical(critVal(penalty = "sqrt", output = "vector", q = ret, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...), critVal(penalty = "sqrt", output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...), info = "sqrt") # log ret <- critVal(penalty = "log", output = "value", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...) expect_identical(critVal(penalty = "log", output = "vector", q = ret, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...), critVal(penalty = "log", output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...), info = "log") # none ret <- critVal(penalty = "none", output = "value", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...) expect_identical(critVal(penalty = "none", output = "vector", q = ret, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...), critVal(penalty = "none", output = "vector", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...), info = "none") # default if (defaultPenalty != "weights") { expect_identical(critVal(output = "value", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...), critVal(penalty = defaultPenalty, output = "value", q = testq, alpha = testalpha, n = testn, nq = testnq, family = testfamily, intervalSystem = testintervalsystem, lengths = testlengths, stat = teststat, r = testr, ...)) } } test_that("function critVal works for given random generated statistics and default values", { testn <- 5L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(stat = teststat)) expect_warning(ret <- critVal(n = testn, stat = teststat)) expect_identical(ret, critVal(alpha = 0.5, n = testn, stat = teststat, nq = testn, family = "gauss", intervalSystem = "all", lengths = 1:testn, penalty = "sqrt", output = "vector")) testVector(teststat = teststat, testn = testn, testlengths = 1:testn, testalpha = 0.1, testweights = rep(0.2, 5), defaultPenalty = "sqrt") testValue(teststat = teststat, testn = testn, testlengths = 1:testn, testalpha = 0.1, testweights = rep(0.2, 5), defaultPenalty = "sqrt") }) test_that("output is tested and works", { testn <- 5L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, output = c("v"))) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, output = "mu")) expect_identical(critVal(alpha = 0.1, n = testn, stat = teststat), critVal(alpha = 0.1, n = testn, stat = teststat, output = "vector")) }) test_that("q is tested and works", { testn <- 5L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(q = "s", alpha = 0.1, n = testn, stat = teststat)) expect_error(critVal(q = Inf, alpha = 0.1, n = testn, stat = teststat)) expect_error(critVal(q = c(rep(1, 4), "s"), alpha = 0.1, n = testn, stat = teststat)) expect_error(critVal(q = c(rep(1, 4), Inf), alpha = 0.1, n = testn, stat = teststat)) expect_error(critVal(q = rep(1, 7), alpha = 0.1, n = testn, stat = teststat)) expect_error(critVal(q = rep(1, 4), alpha = 0.1, n = testn, stat = teststat)) expect_error(critVal(q = rep(1, 5), alpha = 0.1, n = testn, stat = teststat, output = "value")) compare <- 1 attr(compare, "n") <- testn expect_identical(critVal(q = 1, n = testn, stat = teststat, output = "value"), compare) compare <- rep(1, 5) attr(compare, "n") <- testn expect_identical(critVal(q = rep(1, 5), n = testn, stat = teststat), compare) testq <- 1 attr(testq, "n") <- 10L compare <- 1 attr(compare, "n") <- 10L expect_identical(critVal(q = testq, n = testn, stat = teststat, output = "value"), compare) testq <- rep(1, 5) attr(testq, "n") <- 10L compare <- rep(1, 5) attr(compare, "n") <- 10L expect_identical(critVal(q = testq, n = testn, stat = teststat), compare) testq <- 1:10 attr(testq, "n") <- 10L compare <- 1:5 attr(compare, "n") <- 10L expect_identical(critVal(q = testq, n = testn, stat = teststat), compare) }) test_that("alpha is tested and works", { testn <- 30L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(alpha = "s", n = testn, stat = teststat)) expect_error(critVal(alpha = NA, n = testn, stat = teststat)) expect_error(critVal(alpha = Inf, n = testn, stat = teststat)) expect_error(critVal(alpha = c(0.1, 0.05), n = testn, stat = teststat)) expect_error(critVal(alpha = 0, n = testn, stat = teststat)) expect_error(critVal(alpha = 1, n = testn, stat = teststat)) expect_error(critVal(alpha = -0.05, n = testn, stat = teststat)) expect_error(critVal(alpha = 1.1, n = testn, stat = teststat)) testVector(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 0.123, testweights = rep(1 / testn, testn), defaultPenalty = "sqrt") testValue(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 0.123, testweights = rep(1 / testn, testn), defaultPenalty = "sqrt") testVector(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 0.01, testweights = rep(1 / testn, testn), defaultPenalty = "sqrt") testValue(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 0.01, testweights = rep(1 / testn, testn), defaultPenalty = "sqrt") }) test_that("n is tested and works", { testn <- 30L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(n = "s", alpha = 0.1, stat = teststat)) expect_error(critVal(n = c(1L, 2L), alpha = 0.1, stat = teststat)) expect_error(critVal(n = NA, alpha = 0.1, stat = teststat)) expect_error(critVal(n = NULL, alpha = 0.1, stat = teststat)) expect_error(critVal(n = Inf, alpha = 0.1, stat = teststat)) expect_error(critVal(n = 0L, alpha = 0.1, stat = teststat)) expect_error(critVal(n = -1L, alpha = 0.1, stat = teststat)) expect_identical(critVal(n = 30, alpha = 0.1, stat = teststat), critVal(n = 30L, alpha = 0.1, stat = teststat)) expect_identical(critVal(n = 30.5, alpha = 0.1, stat = teststat), critVal(n = 30L, alpha = 0.1, stat = teststat)) }) test_that("nq is tested and works", { testn <- 10L expect_error(critVal(n = testn, nq = "s", alpha = 0.1, q = 1)) expect_error(critVal(n = testn, nq = c(1L, 2L), alpha = 0.1, q = 1)) expect_error(critVal(n = testn, nq = NA, alpha = 0.1, q = 1)) expect_error(critVal(n = testn, nq = Inf, alpha = 0.1, q = 1)) expect_error(critVal(n = testn, nq = NULL, alpha = 0.1, q = 1)) expect_error(critVal(n = testn, nq = 8L, alpha = 0.1, q = 1)) }) test_that("attribute n of q is tested and works", { testn <- 5L teststat <- monteCarloSimulation(n = testn, r = 100L) testq <- 1 attr(testq, "n") <- "s" expect_error(critVal(q = testq, n = testn, stat = teststat, output = "value")) attr(testq, "n") <- c(1L, 2L) expect_error(critVal(q = testq, n = testn, stat = teststat, output = "value")) attr(testq, "n") <- 4L expect_error(critVal(q = testq, n = testn, stat = teststat, output = "value")) attr(testq, "n") <- Inf expect_error(critVal(q = testq, n = testn, stat = teststat, output = "value")) attr(testq, "n") <- NA expect_error(critVal(q = testq, n = testn, stat = teststat, output = "value")) attr(testq, "n") <- 5 expect_identical(critVal(q = testq, n = testn, output = "value"), critVal(q = 1, n = testn, output = "value")) attr(testq, "n") <- 5.5 expect_identical(critVal(q = testq, n = testn, output = "value"), critVal(q = 1, n = testn, output = "value")) attr(testq, "n") <- 10 compare <- 1 attr(compare, "n") <- 10L expect_identical(critVal(q = testq, n = testn, output = "value"), compare) testq <- rep(1, 5) attr(testq, "n") <- "s" expect_error(critVal(q = testq, n = testn, stat = teststat)) attr(testq, "n") <- c(1L, 2L) expect_error(critVal(q = testq, n = testn, stat = teststat)) attr(testq, "n") <- 4L expect_error(critVal(q = testq, n = testn, stat = teststat)) attr(testq, "n") <- Inf expect_error(critVal(q = testq, n = testn, stat = teststat)) attr(testq, "n") <- NA expect_error(critVal(q = testq, n = testn, stat = teststat)) attr(testq, "n") <- 5 expect_identical(critVal(q = testq, n = testn), critVal(q = rep(1, 5), n = testn)) attr(testq, "n") <- 5.5 expect_identical(critVal(q = testq, n = testn), critVal(q = rep(1, 5), n = testn)) attr(testq, "n") <- 10 compare <- rep(1, 5) attr(compare, "n") <- 10L expect_identical(critVal(q = testq, n = testn), compare) testq <- rep(1, 10) attr(testq, "n") <- 10L expect_identical(critVal(q = testq, n = testn), compare) }) test_that("family is tested and works", { testn <- 5L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, family = "")) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, family = 1)) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, family = c("gauss", "hsmuce"))) expect_identical(critVal(n = testn, alpha = 0.1, stat = teststat), critVal(n = testn, alpha = 0.1, stat = teststat, family = "gauss")) }) test_that("intervalSystem is tested and works", { testn <- 5L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, intervalSystem = "")) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, intervalSystem = "dya")) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, intervalSystem = "dyalen")) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, intervalSystem = "dyapar")) expect_identical(critVal(n = testn, alpha = 0.1, stat = teststat), critVal(n = testn, alpha = 0.1, stat = teststat, intervalSystem = "all")) testq <- as.numeric(2^(0:5)) attr(testq, "n") <- 32L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaLen"), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 32L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaLen"), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 64L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaPar"), compare) testq <- as.numeric(1:64) attr(testq, "n") <- 64L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaPar"), compare) testq <- as.numeric(2^(0:6)) attr(testq, "n") <- 64L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaPar"), compare) testn <- 32L teststat <- monteCarloSimulation(n = testn, r = 100L, intervalSystem = "dyaLen") testVector(teststat = teststat, testn = 32L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "dyaLen") testValue(teststat = teststat, testn = 32L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaLen") testn <- 32L teststat <- monteCarloSimulation(n = testn, r = 100L, intervalSystem = "dyaPar") testVector(teststat = teststat, testn = 32L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "dyaPar") testValue(teststat = teststat, testn = 32L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaPar") }) test_that("lengths is tested and works", { testn <- 6L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, lengths = "s")) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, lengths = c(1:5, NA))) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, lengths = c(1:5, Inf))) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, lengths = 0:5)) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, lengths = -1L)) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, lengths = c(1L, 46L))) expect_warning(ret <- critVal(n = testn, alpha = 0.1, stat = teststat, lengths = c(1:6, 6))) expect_identical(ret, critVal(n = testn, alpha = 0.1, stat = teststat, lengths = c(1:6))) expect_identical(critVal(n = testn, alpha = 0.1, stat = teststat, lengths = c(6:1)), critVal(n = testn, alpha = 0.1, stat = teststat, lengths = c(1:6))) expect_identical(critVal(n = testn, alpha = 0.1, stat = teststat, lengths = c(1:6 + 0.5)), critVal(n = testn, alpha = 0.1, stat = teststat, lengths = c(1:6))) testn <- 32L teststat <- monteCarloSimulation(n = testn, r = 100L, intervalSystem = "dyaLen") expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen", lengths = 3L)) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen", lengths = 64L)) testn <- 32L teststat <- monteCarloSimulation(n = testn, r = 100L, intervalSystem = "dyaPar") expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, intervalSystem = "dyaPar", lengths = 3L)) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, intervalSystem = "dyaPar", lengths = 64L)) testq <- as.numeric(1:8) compare <- as.numeric(1:5) attr(compare, "n") <- 8L expect_identical(critVal(q = testq, n = 8L, lengths = 1:5), compare) testq <- as.numeric(3:7) attr(testq, "n") <- 12L compare <- as.numeric(3:7) attr(compare, "n") <- 12L expect_identical(critVal(q = testq, n = 8L, lengths = 3:7), compare) testq <- as.numeric(1:8) attr(testq, "n") <- 12L compare <- as.numeric(3:7) attr(compare, "n") <- 12L expect_identical(critVal(q = testq, n = 8L, lengths = 3:7), compare) testq <- as.numeric(1:12) attr(testq, "n") <- 12L compare <- as.numeric(3:7) attr(compare, "n") <- 12L expect_identical(critVal(q = testq, n = 8L, lengths = 3:7), compare) testq <- as.numeric(1:5) attr(testq, "n") <- 32L compare <- as.numeric(1:5) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 16L, lengths = 2^(0:4), intervalSystem = "dyaPar"), compare) testq <- as.numeric(1:6) attr(testq, "n") <- 32L compare <- as.numeric(1:5) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 16L, lengths = 2^(0:4), intervalSystem = "dyaPar"), compare) testq <- as.numeric(1:5) attr(testq, "n") <- 32L compare <- as.numeric(2:4) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 16L, lengths = 2^(1:3), intervalSystem = "dyaPar"), compare) testq <- as.numeric(1:6) attr(testq, "n") <- 32L compare <- as.numeric(2:4) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 16L, lengths = 2^(1:3), intervalSystem = "dyaLen"), compare) teststat <- monteCarloSimulation(n = 16L, r = 100L) testVector(teststat = teststat, testn = 16L, testlengths = 2^(0:4), compareStat = teststat[2^(0:4), ], testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 5, 5), testintervalsystem = "all") teststat <- monteCarloSimulation(n = 18L, r = 100L) testValue(teststat = teststat, testn = 18L, testlengths = 2^(0:4), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "all") teststat <- monteCarloSimulation(n = 65L, r = 100L, intervalSystem = "dyaLen") testVector(teststat = teststat, testn = 65L, testlengths = c(2L, 8L, 16L, 32L, 64L), testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 5, 5), testintervalsystem = "dyaLen", compareStat = teststat[c(2, 4, 5, 6, 7), ]) testValue(teststat = teststat, testn = 65L, testlengths = c(2L, 8L, 16L, 32L, 64L), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaLen") teststat <- monteCarloSimulation(n = 120L, r = 100L, intervalSystem = "dyaPar") testVector(teststat = teststat, testn = 120L, testlengths = c(1L, 2L, 8L, 32L, 64L), testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 5, 5), testintervalsystem = "dyaPar", compareStat = teststat[c(1, 2, 4, 6, 7), ]) testValue(teststat = teststat, testn = 120L, testlengths = c(1L, 2L, 8L, 32L, 64L), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaPar") }) test_that("penalty is tested and works", { testn <- 10L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(n = testn, penalty = "", alpha = 0.1, stat = teststat)) expect_error(critVal(n = testn, penalty = c("gauss"), alpha = 0.1, stat = teststat)) expect_identical(critVal(n = testn, alpha = 0.1, stat = teststat), critVal(n = testn, penalty = "sqrt", alpha = 0.1, stat = teststat)) }) test_that("weights is tested and works", { testn <- 5L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(n = testn, penalty = "weights", weights = c(rep(1 / 4, 4), "s"), alpha = 0.1, stat = teststat)) expect_error(critVal(n = testn, penalty = "weights", weights = c(rep(1 / 4, 4), NA), alpha = 0.1, stat = teststat)) expect_error(critVal(n = testn, penalty = "weights", weights = c(rep(1 / 4, 4), Inf), alpha = 0.1, stat = teststat)) expect_error(critVal(n = testn, penalty = "weights", weights = c(rep(1 / 4, 4)), alpha = 0.1, stat = teststat)) expect_error(critVal(n = testn, penalty = "weights", weights = c(rep(1 / 4, 4), 0), alpha = 0.1, stat = teststat)) expect_error(critVal(n = testn, penalty = "weights", weights = c(rep(1 / 4, 4), -0.1), alpha = 0.1, stat = teststat)) expect_error(critVal(n = testn, penalty = "weights", weights = rep(1 / 8, 8), alpha = 0.1, stat = teststat)) expect_error(critVal(n = testn, penalty = "weights", weights = rep(1 / 5, 5), lengths = 1:4, alpha = 0.1, stat = teststat)) expect_identical(critVal(n = testn, alpha = 0.1, penalty = "weights", stat = teststat), critVal(n = testn, alpha = 0.1, penalty = "weights", weights = rep(1 / testn, testn), stat = teststat)) expect_identical(critVal(n = testn, penalty = "weights", weights = rep(1, testn), alpha = 0.1, stat = teststat), critVal(n = testn, penalty = "weights", weights = rep(1 / testn, testn), alpha = 0.1, stat = teststat)) expect_warning(ret <- critVal(n = testn, weights = rep(1 / testn, testn), alpha = 0.1, stat = teststat)) expect_identical(ret, critVal(n = testn, alpha = 0.1, stat = teststat)) testVector(teststat = teststat, testn = testn, testnq = 20L, testlengths = 1:testn, testalpha = 0.01, compareWeights = rep(1 / testn, testn), defaultPenalty = "sqrt") testValue(teststat = teststat, testn = testn, testnq = 20L, testlengths = 1:testn, testalpha = 0.01, defaultPenalty = "sqrt") testVector(teststat = teststat, testn = testn, testnq = 20L, testlengths = 1:testn, testalpha = 0.01, testweights = 5:1 / sum(5:1), defaultPenalty = "sqrt") testValue(teststat = teststat, testn = testn, testnq = 20L, testlengths = 1:testn, testalpha = 0.01, testweights = 5:1 / sum(5:1), defaultPenalty = "sqrt") testVector(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 0.01, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "sqrt") testValue(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 0.01, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "sqrt") testVector(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 1e-6, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "sqrt") testValue(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 1e-6, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "sqrt") testVector(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 1 - 1e-6, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "sqrt") testValue(teststat = teststat, testn = testn, testnq = testn, testlengths = 1:testn, testalpha = 1- 1e-6, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "sqrt") }) test_that("stat is tested and works if it is a matrix", { testn <- 6L teststat <- matrix(abs(rnorm(1e4 * testn)), testn, 1e4) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(n = 10L, r = 100L, intervalSystem = "dyaLen") attr(teststat, "n") <- 11L expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen")) teststat <- monteCarloSimulation(n = 10L, r = 100L, family = "hsmuce") expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(n = 10L, r = 100L, family = "hsmuce") expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, penalty = "weights")) teststat <- monteCarloSimulation(n = 10L, r = 100L, intervalSystem = "dyaPar") expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen")) teststat <- monteCarloSimulation(n = 10L, r = 100L) testVector(teststat = teststat, testn = 6L, testlengths = 1:6, compareStat = teststat[1:6, ], testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "all", nstat = 10L) testValue(teststat = teststat, testn = 6L, testlengths = 1:6, testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "all") teststat <- monteCarloSimulation(n = 40L, r = 100L, intervalSystem = "dyaLen") testVector(teststat = teststat, testn = 32L, testlengths = 2^(0:5), compareStat = teststat[1:6, ], testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "dyaLen", nstat = 40L) testValue(teststat = teststat, testn = 32L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaLen") teststat <- monteCarloSimulation(n = 64L, r = 100L, intervalSystem = "dyaPar") testVector(teststat = teststat, testn = 32L, testlengths = 2^(0:5), compareStat = teststat[1:6, ], testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "dyaPar", nstat = 64L) testValue(teststat = teststat, testn = 32L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaPar") teststat <- monteCarloSimulation(n = 10L, r = 100L) testVector(teststat = teststat, testn = 8L, testlengths = 1:6, compareStat = teststat[1:6, ], testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "all", nstat = 10L) testValue(teststat = teststat, testn = 8L, testlengths = 1:6, testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "all") teststat <- monteCarloSimulation(n = 40L, r = 100L, intervalSystem = "dyaLen") testVector(teststat = teststat, testn = 40L, testlengths = 2^(0:5), compareStat = teststat[1:6, ], testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "dyaLen", nstat = 40L) testValue(teststat = teststat, testn = 40L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaLen") teststat <- monteCarloSimulation(n = 127L, r = 100L, intervalSystem = "dyaPar") testVector(teststat = teststat, testn = 65L, testlengths = 2^(0:5), compareStat = teststat[1:6, ], testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "dyaPar", nstat = 127L) testValue(teststat = teststat, testn = 65L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaPar") }) test_that("output is tested", { testn <- 6L teststat <- monteCarloSimulation(n = testn, r = 100L) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, output = 1)) expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, output = "mu")) expect_identical(critVal(n = testn, alpha = 0.1, stat = teststat), critVal(n = testn, alpha = 0.1, stat = teststat, output = "ve")) }) test_that("stat is tested and works if it is a vector", { teststat <- 1 + rnorm(1e4) expect_error(critVal(n = 32L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(n = 10L, r = 100L, intervalSystem = "dyaLen", output = "maximum") attr(teststat, "n") <- 11L expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen")) teststat <- monteCarloSimulation(n = 10L, r = 100L, family = "hsmuce", output = "maximum") expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(n = 10L, r = 100L, intervalSystem = "dyaPar", output = "maximum") expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen")) teststat <- monteCarloSimulation(n = 10L, r = 100L, lengths = 3:5, output = "maximum") expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, lengths = 4:6)) teststat <- monteCarloSimulation(n = 10L, r = 100L, penalty = "log", output = "maximum") expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(n = 10L, r = 100L, output = "maximum") expect_error(critVal(n = 10L, alpha = 0.1, penalty = "weights", stat = teststat)) teststatmatrix <- monteCarloSimulation(n = 6L, r = 100L) teststatvector <- monteCarloSimulation(n = 6L, r = 100L, output = "maximum") expect_identical(critVal(n = 6L, alpha = 0.1, stat = teststatmatrix), critVal(n = 6L, alpha = 0.1, stat = teststatvector)) teststatmatrix <- monteCarloSimulation(n = 15L, r = 100L) teststatvector <- monteCarloSimulation(n = 15L, r = 100L, output = "maximum", penalty = "log", lengths = 1:12) expect_identical(critVal(n = 12L, alpha = 0.1, stat = teststatmatrix, penalty = "log"), critVal(n = 12L, alpha = 0.1, stat = teststatvector, penalty = "log")) teststatmatrix <- monteCarloSimulation(n = 32L, r = 100L, intervalSystem = "dyaLen") teststatvector <- monteCarloSimulation(n = 32L, r = 100L, intervalSystem = "dyaLen", output = "maximum", penalty = "none", lengths = 2^(0:3)) expect_identical(critVal(n = 8L, alpha = 0.1, stat = teststatmatrix, penalty = "none", intervalSystem = "dyaLen"), critVal(n = 8L, alpha = 0.1, stat = teststatvector, penalty = "none", intervalSystem = "dyaLen")) teststatmatrix <- monteCarloSimulation(n = 32L, r = 100L, intervalSystem = "dyaLen") teststatvector <- monteCarloSimulation(n = 32L, r = 100L, intervalSystem = "dyaLen", output = "maximum", penalty = "sqrt", lengths = c(1, 2, 8, 16)) expect_identical(critVal(n = 32L, alpha = 0.1, stat = teststatmatrix, penalty = "sqrt", lengths = c(1, 2, 8, 16), intervalSystem = "dyaLen"), critVal(n = 32L, alpha = 0.1, stat = teststatvector, penalty = "sqrt", lengths = c(1, 2, 8, 16), intervalSystem = "dyaLen")) teststatmatrix <- monteCarloSimulation(n = 32L, r = 100L, intervalSystem = "dyaPar") teststatvector <- monteCarloSimulation(n = 32L, r = 100L, intervalSystem = "dyaPar", output = "maximum", penalty = "log", lengths = c(2, 4, 8, 16)) expect_identical(critVal(n = 32L, alpha = 0.1, stat = teststatmatrix, penalty = "log", lengths = c(2, 4, 8, 16), intervalSystem = "dyaPar"), critVal(n = 32L, alpha = 0.1, stat = teststatvector, penalty = "log", lengths = c(2, 4, 8, 16), intervalSystem = "dyaPar")) teststatmatrix <- monteCarloSimulation(n = 32L, r = 100L, intervalSystem = "dyaPar") teststatvector <- monteCarloSimulation(n = 32L, r = 100L, intervalSystem = "dyaPar", output = "maximum", penalty = "none", lengths = 8) expect_identical(critVal(n = 32L, alpha = 0.1, stat = teststatmatrix, penalty = "none", lengths = 8, intervalSystem = "dyaPar"), critVal(n = 32L, alpha = 0.1, stat = teststatvector, penalty = "none", lengths = 8, intervalSystem = "dyaPar")) }) test_that("r is tested and works", { expect_error(critVal(100L, alpha = 0.1, r = "s", options = list(load = list()))) expect_error(critVal(100L, alpha = 0.1, r = 0, options = list(load = list()))) expect_error(critVal(100L, alpha = 0.1, r = c(100, 200), options = list(load = list()))) expect_identical(critVal(10L, alpha = 0.1, r = 100.5, options = list(load = list())), critVal(10L, alpha = 0.1, r = 100L, options = list(load = list()))) teststat <- monteCarloSimulation(n = 3L) expect_identical(critVal(3, alpha = 0.1, options = list(load = list())), critVal(3, alpha = 0.1, stat = teststat)) }) test_that("stat works if it has to be simulated", { expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulat = "vector", save = list(), load = list()))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = "vector")) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vecto", save = list(), load = list()))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = c("vector", "matrix"), save = list(), load = list()))) expect_error(critVal(100L, alpha = 0.1, r = 100L, penalty = "weights", options = list(simulation = c("vector"), save = list(), load = list()))) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(150L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, nq = 150L, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrix", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) teststat <- monteCarloSimulation(127L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrixIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(127L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) teststat <- monteCarloSimulation(118L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", nq = 118L, options = list(simulation = "matrixIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) teststat <- monteCarloSimulation(127L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum", intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log") expect_identical(critVal(100L, alpha = 0.1, r = 100L, intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log", options = list(simulation = "vector", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log")) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 2:30) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = 2:30, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = 2:30)) teststat <- monteCarloSimulation(120L, r = 100L, output = "maximum", lengths = 2:30) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = 2:30, nq = 120L, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = 2:30)) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, lengths = c(2, 7:8, 100), options = list(simulation = "matrix", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, lengths = c(2, 7:8, 100))) teststat <- monteCarloSimulation(127L, r = 100L, intervalSystem = "dyaLen") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 32, 64), options = list(simulation = "matrixIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 32, 64))) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum", intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log") expect_identical(critVal(100L, alpha = 0.1, r = 100L, intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log", options = list(simulation = "vector", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log")) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 2:30) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = 2:30, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = 2:30)) teststat <- monteCarloSimulation(120L, r = 100L, output = "maximum", lengths = 2:30) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = 2:30, nq = 120L, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = 2:30)) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, lengths = c(2, 7:8, 100), options = list(simulation = "matrix", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, lengths = c(2, 7:8, 100))) teststat <- monteCarloSimulation(127L, r = 100L, intervalSystem = "dyaLen") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 32, 64), options = list(simulation = "matrixIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 32, 64))) }) test_that("RDSfile is tested and stat can be saved in / loaded from an RDS file", { testfile <- c(tempfile(pattern = "file", tmpdir = tempdir(), fileext = ".RDS"), tempfile(pattern = "file", tmpdir = tempdir(), fileext = ".RDS")) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(RDSfile = testfile, test = 1), load = list()))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(RDSfile = testfile), load = list(test = "test")))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(RDSfile = c(testfile, testfile)), load = list()))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(RDSfile = 10), load = list()))) teststat <- monteCarloSimulation(100L, r = 100L) teststatvec <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(readRDS(testfile[1]), teststatvec) expect_identical(readRDS(testfile[2]), teststat) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", load = list(RDSfile = 10), save = list()))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", load = list(RDSfile = testfile), save = list()))) unlink(testfile) teststat <- monteCarloSimulation(n = 10L, r = 100L, intervalSystem = "dyaLen") attr(teststat, "n") <- 11L saveRDS(teststat, file = testfile[1]) expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen", options = list(load = list(RDSfile = testfile[1])))) teststat <- monteCarloSimulation(n = 10L, r = 100L, family = "hsmuce") saveRDS(teststat, file = testfile[1]) expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, options = list(load = list(RDSfile = testfile[1])))) teststat <- monteCarloSimulation(n = 10L, r = 100L, family = "hsmuce") saveRDS(teststat, file = testfile[1]) expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, penalty = "weights", options = list(load = list(RDSfile = testfile[1])))) teststat <- monteCarloSimulation(n = 10L, r = 100L, intervalSystem = "dyaPar") saveRDS(teststat, file = testfile[1]) expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen", options = list(load = list(RDSfile = testfile[1])))) teststat <- monteCarloSimulation(n = 10L, r = 100L, intervalSystem = "dyaLen", output = "maximum") attr(teststat, "n") <- 11L saveRDS(teststat, file = testfile[1]) expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen", options = list(load = list(RDSfile = testfile[1])))) teststat <- monteCarloSimulation(n = 10L, r = 100L, family = "hsmuce", output = "maximum") saveRDS(teststat, file = testfile[1]) expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, options = list(load = list(RDSfile = testfile[1])))) teststat <- monteCarloSimulation(n = 10L, r = 100L, intervalSystem = "dyaPar", output = "maximum") saveRDS(teststat, file = testfile[1]) expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, intervalSystem = "dyaLen", options = list(load = list(RDSfile = testfile[1])))) teststat <- monteCarloSimulation(n = 10L, r = 100L, lengths = 3:5, output = "maximum") saveRDS(teststat, file = testfile[1]) expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, lengths = 4:6, options = list(load = list(RDSfile = testfile[1])))) teststat <- monteCarloSimulation(n = 10L, r = 100L, penalty = "log", output = "maximum") saveRDS(teststat, file = testfile[1]) expect_error(critVal(n = 10L, alpha = 0.1, stat = teststat, options = list(load = list(RDSfile = testfile[1])))) unlink(testfile[1]) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(readRDS(testfile[1]), teststat) expect_false(file.exists(testfile[2])) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(readRDS(testfile[1]), teststat) expect_false(file.exists(testfile[2])) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "vector", penalty = "weights", options = list(simulation = "matrix", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", penalty = "weights")) expect_false(file.exists(testfile[1])) expect_identical(readRDS(testfile[2]), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[2]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(100L, r = 100L) teststatvec <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(readRDS(testfile[1]), teststatvec) expect_identical(readRDS(testfile[2]), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[2]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(127L, r = 100L) teststatvec <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrixIncreased", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(readRDS(testfile[1]), teststatvec) expect_identical(readRDS(testfile[2]), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[2]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) testfile <- c("", tempfile(pattern = "file", tmpdir = tempdir(), fileext = ".RDS")) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) expect_false(file.exists(testfile[2])) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_false(file.exists(testfile[2])) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(readRDS(testfile[2]), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[2]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(127L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrixIncreased", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(readRDS(testfile[2]), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[2]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) testfile <- c(tempfile(pattern = "file", tmpdir = tempdir(), fileext = ".RDS"), "") teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(readRDS(testfile[1]), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(readRDS(testfile[1]), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(100L, r = 100L) teststatvec <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(readRDS(testfile[1]), teststatvec) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(127L, r = 100L) teststatvec <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrixIncreased", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(readRDS(testfile[1]), teststatvec) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(RDSfile = c("", "")), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) testfile <- tempfile(pattern = "file", tmpdir = tempdir(), fileext = ".RDS") teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(readRDS(testfile), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(readRDS(testfile), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(100L, r = 100L) teststatvec <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(readRDS(testfile), teststatvec) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(127L, r = 100L) teststatvec <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrixIncreased", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(readRDS(testfile), teststatvec) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile[1]))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "vector", penalty = "weights", options = list(simulation = "matrix", save = list(RDSfile = testfile), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", penalty = "weights")) expect_identical(readRDS(testfile), teststat) expect_identical(critVal(100L, alpha = 0.1, r = 50L, options = list(simulation = "vector", save = list(), load = list(RDSfile = testfile))), critVal(100L, alpha = 0.1, stat = teststat)) unlink(testfile) }) test_that("variable is tested and stat can be saved in a variable", { testvariable <- c("testsavevector", "testsavematrix") testStepR <- new.env() expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(variable = c(testvariable, testvariable)), load = list(), envir = testStepR))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(variable = 10), load = list(), envir = testStepR))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(variable = testvariable), load = list(), envir = "testStepR"))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(variable = testvariable), load = list(), envir = c(testStepR, testStepR)))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(variable = testvariable), load = list(), envir = c(testStepR, 10)))) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(get("testsavevector", envir = testStepR), teststat) expect_false(exists("testsavematrix", envir = testStepR)) remove(testsavevector, envir = testStepR) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(get("testsavevector", envir = testStepR), teststat) expect_false(exists("testsavematrix", envir = testStepR)) remove(testsavevector, envir = testStepR) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "vector", penalty = "weights", options = list(simulation = "matrix", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", penalty = "weights")) expect_identical(get("testsavematrix", envir = testStepR), teststat) expect_false(exists("testsavevector", envir = testStepR)) remove(testsavematrix, envir = testStepR) teststat <- monteCarloSimulation(100L, r = 100L) teststatvec <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(get("testsavematrix", envir = testStepR), teststat) expect_identical(get("testsavevector", envir = testStepR), teststatvec) remove(testsavematrix, envir = testStepR) remove(testsavevector, envir = testStepR) teststat <- monteCarloSimulation(127L, r = 100L) teststatvec <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrixIncreased", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(get("testsavematrix", envir = testStepR), teststat) expect_identical(get("testsavevector", envir = testStepR), teststatvec) remove(testsavematrix, envir = testStepR) remove(testsavevector, envir = testStepR) testvariable <- c("", "testsavematrix") teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat)) expect_false(exists("testsavevector", envir = testStepR)) expect_false(exists("testsavematrix", envir = testStepR)) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_false(exists("testsavevector", envir = testStepR)) expect_false(exists("testsavematrix", envir = testStepR)) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(get("testsavematrix", envir = testStepR), teststat) expect_false(exists("testsavevector", envir = testStepR)) remove(testsavematrix, envir = testStepR) teststat <- monteCarloSimulation(127L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrixIncreased", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(get("testsavematrix", envir = testStepR), teststat) expect_false(exists("testsavevector", envir = testStepR)) remove(testsavematrix, envir = testStepR) testvariable <- c("testsavevector", "") teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(get("testsavevector", envir = testStepR), teststat) expect_false(exists("testsavematrix", envir = testStepR)) remove(testsavevector, envir = testStepR) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(get("testsavevector", envir = testStepR), teststat) expect_false(exists("testsavematrix", envir = testStepR)) remove(testsavevector, envir = testStepR) teststat <- monteCarloSimulation(100L, r = 100L) teststatvec <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(get("testsavevector", envir = testStepR), teststatvec) expect_false(exists("testsavematrix", envir = testStepR)) remove(testsavevector, envir = testStepR) teststat <- monteCarloSimulation(127L, r = 100L) teststatvec <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrixIncreased", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(get("testsavevector", envir = testStepR), teststatvec) expect_false(exists("testsavematrix", envir = testStepR)) remove(testsavevector, envir = testStepR) testvariable <- "testsave" teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "vector", penalty = "weights", options = list(simulation = "matrix", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", penalty = "weights")) expect_identical(get("testsave", envir = testStepR), teststat) remove(testsave, envir = testStepR) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(get("testsave", envir = testStepR), teststat) remove(testsave, envir = testStepR) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(get("testsave", envir = testStepR), teststat) remove(testsave, envir = testStepR) teststat <- monteCarloSimulation(100L, r = 100L) teststatvec <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_identical(get("testsave", envir = testStepR), teststatvec) remove(testsave, envir = testStepR) teststat <- monteCarloSimulation(127L, r = 100L) teststatvec <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "matrixIncreased", save = list(variable = testvariable), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat)) expect_identical(get("testsave", envir = testStepR), teststatvec) remove(testsave, envir = testStepR) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(RDSfile = c("", "")), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value")) expect_false(exists("testsavevector", envir = testStepR)) expect_false(exists("testsavematrix", envir = testStepR)) }) test_that("stat can be saved in and loaded from the workspace", { testStepR <- new.env() expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(workspace = "matri"), load = list(), envir = testStepR))) expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(workspace = c("vector", "matri")), load = list(), envir = testStepR))) expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", load = list(workspace = "matri"), save = list(), envir = testStepR))) expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", load = list(workspace = c("vector", "matri")), save = list(), envir = testStepR))) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(workspace = "matrix"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrix", save = list(), load = list(workspace = "matrix"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) remove(critValStepRTab, envir = testStepR) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(workspace = "matrixIncreased"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(exists("critValStepRTab", where = testStepR, inherits = FALSE)) teststat <- monteCarloSimulation(127L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(workspace = "matrixIncreased"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrixIncreased", save = list(), load = list(workspace = "matrixIncreased"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) remove(critValStepRTab, envir = testStepR) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(workspace = "matrix"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(exists("critValStepRTab", where = testStepR, inherits = FALSE)) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(workspace = "vector"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "vector", save = list(), load = list(workspace = "vector"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) remove(critValStepRTab, envir = testStepR) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(workspace = "vectorIncreased"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(exists("critValStepRTab", where = testStepR, inherits = FALSE)) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(workspace = "vectorIncreased"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "vectorIncreased", save = list(), load = list(workspace = "vectorIncreased"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) remove(critValStepRTab, envir = testStepR) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(workspace = "vector"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(exists("critValStepRTab", where = testStepR, inherits = FALSE)) teststat <- monteCarloSimulation(100L, r = 100L) teststatvec <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(workspace = "vector"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststatvec) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrix", save = list(), load = list(workspace = "vector"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) remove(critValStepRTab, envir = testStepR) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(workspace = "vectorIncreased"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(exists("critValStepRTab", where = testStepR, inherits = FALSE)) teststat <- monteCarloSimulation(127L, r = 100L) teststatvec <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(workspace = "vectorIncreased"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststatvec) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrixIncreased", save = list(), load = list(workspace = "vectorIncreased"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) remove(critValStepRTab, envir = testStepR) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(workspace = "vector"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(exists("critValStepRTab", where = testStepR, inherits = FALSE)) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(workspace = "matrix"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(exists("critValStepRTab", where = testStepR, inherits = FALSE)) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(workspace = c("matrix", "vectorIncreased")), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrix", save = list(), load = list(workspace = "matrix"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) remove(critValStepRTab, envir = testStepR) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(workspace = c("matrix", "vectorIncreased")), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrix", save = list(), load = list(workspace = "vectorIncreased"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) remove(critValStepRTab, envir = testStepR) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(workspace = c("matrix", "vector")), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(exists("critValStepRTab", where = testStepR, inherits = FALSE)) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(workspace = c("vector")), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) teststat200 <- monteCarloSimulation(100L, r = 200L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 200L, output = "value", options = list(simulation = "vector", save = list(workspace = c("vector")), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat200, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat200) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(workspace = c("vector")), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat200) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "vector", save = list(), load = list(workspace = "vector"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat200, output = "value", options = list())) remove(critValStepRTab, envir = testStepR) teststat1 <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(workspace = c("matrixIncreased", "vector", "matrix")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat1, output = "value", options = list())) teststat2 <- monteCarloSimulation(98L, r = 100L) teststat3 <- monteCarloSimulation(98L, r = 100L, output = "maximum") expect_identical(critVal(98L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(workspace = c("matrixIncreased", "vector", "matrix")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(98L, alpha = 0.1, stat = teststat2, output = "value", options = list())) teststat <- monteCarloSimulation(127L, r = 100L) teststat4 <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:102) expect_identical(critVal(102L, alpha = 0.1, r = 100L, output = "value", stat = teststat, options = list(simulation = "matrixIncreased", save = list(workspace = c("matrix", "vectorIncreased")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(102L, alpha = 0.1, stat = teststat, output = "value", options = list())) teststat1 <- monteCarloSimulation(100L, r = 200L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 200L, output = "value", options = list(simulation = "vector", save = list(workspace = c("matrixIncreased", "vector", "matrix")), load = list(workspace = c("vector", "matrix")), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat1, output = "value", options = list())) expect_identical(critVal(102L, alpha = 0.1, r = 100L, output = "vector", options = list(simulation = "matrix", save = list(workspace = c("matrix", "vectorIncreased")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(102L, alpha = 0.1, stat = teststat4, output = "vector", options = list())) teststat5 <- monteCarloSimulation(73L, r = 100L, output = "maximum") expect_identical(critVal(73L, alpha = 0.1, r = 100L, output = "vector", stat = teststat5, options = list(simulation = "vector", save = list(workspace = c("matrix", "vectorIncreased")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(73L, alpha = 0.1, stat = teststat5, output = "vector", options = list())) teststat6 <- monteCarloSimulation(127L, r = 100L) expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", options = list(simulation = "matrixIncreased", save = list(workspace = c("matrixIncreased", "vector")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", options = list())) expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", options = list(simulation = "matrixIncreased", save = list(workspace = c("matrixIncreased", "vector")), load = list(workspace = c("matrixIncreased", "matrix")), envir = testStepR)), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", options = list())) teststat7 <- monteCarloSimulation(74L, r = 200L, output = "maximum") expect_identical(critVal(74L, alpha = 0.1, r = 200L, output = "vector", options = list(simulation = "vector", save = list(workspace = c("matrix", "vectorIncreased", "vector")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(74L, alpha = 0.1, stat = teststat7, output = "vector", options = list())) teststat8 <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:120) expect_identical(critVal(120L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(workspace = c("matrixIncreased", "vectorIncreased", "vector")), load = list(workspace = c("vectorIncreased", "matrixIncreased", "vector")), envir = testStepR)), critVal(120L, alpha = 0.1, stat = teststat8, output = "value", options = list())) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 7L) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat1) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[2]], teststat2) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[3]], teststat3) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[4]], teststat4) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[5]], teststat6) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[6]], teststat7) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[7]], teststat8) remove(critValStepRTab, envir = testStepR) teststat1 <- monteCarloSimulation(100L, r = 100L, output = "maximum", lengths = c(23:34, 45:67)) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = c(23:34, 45:67), options = list(simulation = "vector", save = list(workspace = c("matrixIncreased", "vector", "matrix")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat1, output = "value", lengths = c(23:34, 45:67), options = list())) teststat2 <- monteCarloSimulation(98L, r = 100L, intervalSystem = "dyaPar") teststat3 <- monteCarloSimulation(98L, r = 100L, output = "maximum", intervalSystem = "dyaPar", lengths = 2^(1:3)) expect_identical(critVal(98L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaPar", lengths = 2^(1:3), options = list(simulation = "matrix", save = list(workspace = c("matrixIncreased", "vector", "matrix")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(98L, alpha = 0.1, stat = teststat2, output = "value", intervalSystem = "dyaPar", lengths = 2^(1:3), options = list())) teststat <- monteCarloSimulation(127L, r = 100L) teststat4 <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:102, penalty = "log") expect_identical(critVal(102L, alpha = 0.1, r = 100L, output = "value", stat = teststat, penalty = "log", options = list(simulation = "matrixIncreased", save = list(workspace = c("matrix", "vectorIncreased")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(102L, alpha = 0.1, stat = teststat, output = "value", penalty = "log", options = list())) teststat1 <- monteCarloSimulation(100L, r = 200L, output = "maximum", lengths = c(23:34, 45:67)) expect_identical(critVal(100L, alpha = 0.1, r = 200L, output = "value", lengths = c(23:34, 45:67), options = list(simulation = "vector", save = list(workspace = c("matrixIncreased", "vector", "matrix")), load = list(workspace = c("vector", "matrix")), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat1, output = "value", lengths = c(23:34, 45:67), options = list())) expect_identical(critVal(102L, alpha = 0.1, r = 100L, output = "vector", lengths = 1:102, penalty = "log", options = list(simulation = "matrix", save = list(workspace = c("matrix", "vectorIncreased")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(102L, alpha = 0.1, stat = teststat4, output = "vector", lengths = 1:102, penalty = "log", options = list())) teststat5 <- monteCarloSimulation(73L, r = 100L, output = "maximum", intervalSystem = "dyaLen", lengths = 16L, penalty = "log") expect_identical(critVal(73L, alpha = 0.1, r = 100L, output = "vector", intervalSystem = "dyaLen", lengths = 16L, penalty = "log", stat = teststat5, options = list(simulation = "vector", save = list(workspace = c("matrix", "vectorIncreased")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(73L, alpha = 0.1, stat = teststat5, output = "vector", intervalSystem = "dyaLen", lengths = 16L, penalty = "log", options = list())) teststat6 <- monteCarloSimulation(125L, r = 100L, intervalSystem = "dyaLen") expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", penalty = "weights", nq = 125L, weights = 1:7 / sum(1:7), intervalSystem = "dyaLen", options = list(simulation = "matrixIncreased", save = list(workspace = c("matrixIncreased", "vector")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR)), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", penalty = "weights", weights = 1:7 / sum(1:7), intervalSystem = "dyaLen", options = list())) expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", penalty = "log", intervalSystem = "dyaLen", lengths = 2^(4:6), nq = 125L, options = list(simulation = "matrix", save = list(workspace = c("matrixIncreased", "vector")), load = list(workspace = c("matrixIncreased", "matrix")), envir = testStepR)), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", penalty = "log", intervalSystem = "dyaLen", lengths = 2^(4:6), options = list())) teststat <- monteCarloSimulation(74L, r = 200L, output = "maximum", rand.gen = function(data) {rnorm(74)}) expect_identical(critVal(74L, alpha = 0.1, r = 200L, output = "vector", rand.gen = function(data) {rnorm(74)}, options = list(simulation = "vector", save = list(workspace = c("matrix", "vector")), load = list(workspace = c("vector", "matrix")), envir = testStepR)), critVal(74L, alpha = 0.1, stat = teststat, output = "vector", options = list())) teststat7 <- monteCarloSimulation(125L, r = 100L, output = "maximum", intervalSystem = "dyaLen", lengths = 2^(3:5), penalty = "log") expect_identical(critVal(120L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", lengths = 2^(3:5), penalty = "log", nq = 125L, options = list(simulation = "vector", save = list(workspace = c("matrixIncreased", "vectorIncreased", "vector")), load = list(workspace = c("vectorIncreased", "matrixIncreased", "vector")), envir = testStepR)), critVal(120L, alpha = 0.1, stat = teststat7, output = "value", options = list(), intervalSystem = "dyaLen", lengths = 2^(3:5), penalty = "log")) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 6L) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat1) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[2]], teststat2) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[3]], teststat3) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[4]], teststat4) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[5]], teststat6) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[6]], teststat7) remove(critValStepRTab, envir = testStepR) }) test_that("stat can be saved in and loaded from the file system", { teststat <- monteCarloSimulation(100L, r = 100L) expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = "matri"), load = list(), dirs = "testStepR"))) expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = c("vector", "matri")), load = list(), dirs = "testStepR"))) expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", load = list(fileSystem = "matri"), save = list(), dirs = "testStepR"))) expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", load = list(fileSystem = c("vector", "matri")), save = list(), dirs = "testStepR"))) expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = "matrix"), load = list(), dirs = c("testStepR", "test")))) expect_error(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = "matrix"), load = list(), dirs = 10))) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = "matrix"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststat, "keyList"), dirs = "testStepR"), teststat) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrix", save = list(), load = list(fileSystem = "matrix"), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = "matrixIncreased"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) teststat <- monteCarloSimulation(127L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(fileSystem = "matrixIncreased"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststat, "keyList"), dirs = "testStepR"), teststat) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrixIncreased", save = list(), load = list(fileSystem = "matrixIncreased"), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(fileSystem = "matrix"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(fileSystem = "vector"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststat, "keyList"), dirs = "testStepR"), teststat) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "vector", save = list(), load = list(fileSystem = "vector"), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(fileSystem = "vectorIncreased"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(fileSystem = "vectorIncreased"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststat, "keyList"), dirs = "testStepR"), teststat) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "vectorIncreased", save = list(), load = list(fileSystem = "vectorIncreased"), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(fileSystem = "vector"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) teststat <- monteCarloSimulation(100L, r = 100L) teststatvec <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = "vector"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststatvec, "keyList"), dirs = "testStepR"), teststatvec) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrix", save = list(), load = list(fileSystem = "vector"), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = "vectorIncreased"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) teststat <- monteCarloSimulation(127L, r = 100L) teststatvec <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(fileSystem = "vectorIncreased"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststatvec, "keyList"), dirs = "testStepR"), teststatvec) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrixIncreased", save = list(), load = list(fileSystem = "vectorIncreased"), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(fileSystem = "vector"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(fileSystem = "matrix"), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) teststat <- monteCarloSimulation(100L, r = 100L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststat, "keyList"), dirs = "testStepR"), teststat) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrix", save = list(), load = list(fileSystem = "matrix"), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:100) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vectorIncreased", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststat, "keyList"), dirs = "testStepR"), teststat) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "matrix", save = list(), load = list(fileSystem = "vectorIncreased"), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", save = list(fileSystem = c("matrix", "vector")), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(fileSystem = c("vector")), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststat, "keyList"), dirs = "testStepR"), teststat) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) teststat200 <- monteCarloSimulation(100L, r = 200L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 200L, output = "value", options = list(simulation = "vector", save = list(fileSystem = c("vector")), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat200, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststat200, "keyList"), dirs = "testStepR"), teststat200) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(fileSystem = c("vector")), load = list(), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat, output = "value", options = list())) expect_identical(R.cache::loadCache(attr(teststat200, "keyList"), dirs = "testStepR"), teststat200) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "value", options = list(simulation = "vector", save = list(), load = list(fileSystem = "vector"), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat200, output = "value", options = list())) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) teststat1 <- monteCarloSimulation(100L, r = 100L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(fileSystem = c("matrixIncreased", "vector", "matrix")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat1, output = "value", options = list())) teststat2 <- monteCarloSimulation(98L, r = 100L) teststat3 <- monteCarloSimulation(98L, r = 100L, output = "maximum") expect_identical(critVal(98L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrix", save = list(fileSystem = c("matrixIncreased", "vector", "matrix")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(98L, alpha = 0.1, stat = teststat2, output = "value", options = list())) teststat <- monteCarloSimulation(127L, r = 100L) teststat4 <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:102) expect_identical(critVal(102L, alpha = 0.1, r = 100L, output = "value", stat = teststat, options = list(simulation = "matrixIncreased", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(102L, alpha = 0.1, stat = teststat, output = "value", options = list())) teststat1 <- monteCarloSimulation(100L, r = 200L, output = "maximum") expect_identical(critVal(100L, alpha = 0.1, r = 200L, output = "value", options = list(simulation = "vector", save = list(fileSystem = c("matrixIncreased", "vector", "matrix")), load = list(fileSystem = c("vector", "matrix")), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat1, output = "value", options = list())) expect_identical(critVal(102L, alpha = 0.1, r = 100L, output = "vector", options = list(simulation = "matrix", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(102L, alpha = 0.1, stat = teststat4, output = "vector", options = list())) teststat5 <- monteCarloSimulation(73L, r = 100L, output = "maximum") expect_identical(critVal(73L, alpha = 0.1, r = 100L, output = "vector", stat = teststat5, options = list(simulation = "vector", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(73L, alpha = 0.1, stat = teststat5, output = "vector", options = list())) teststat6 <- monteCarloSimulation(127L, r = 100L) expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", options = list(simulation = "matrixIncreased", save = list(fileSystem = c("matrixIncreased", "vector")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", options = list())) expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", options = list(simulation = "matrixIncreased", save = list(fileSystem = c("matrixIncreased", "vector")), load = list(fileSystem = c("matrixIncreased", "matrix")), dirs = "testStepR")), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", options = list())) teststat7 <- monteCarloSimulation(74L, r = 200L, output = "maximum") expect_identical(critVal(74L, alpha = 0.1, r = 200L, output = "vector", options = list(simulation = "vector", save = list(fileSystem = c("matrix", "vector")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(74L, alpha = 0.1, stat = teststat7, output = "vector", options = list())) teststat8 <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:120) expect_identical(critVal(120L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "vector", save = list(fileSystem = c("matrixIncreased", "vector", "vectorIncreased")), load = list(fileSystem = c("vectorIncreased", "matrixIncreased", "vector")), dirs = "testStepR")), critVal(120L, alpha = 0.1, stat = teststat8, output = "value", options = list())) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 7L) expect_identical(R.cache::loadCache(attr(teststat1, "keyList"), dirs = "testStepR"), teststat1) expect_identical(R.cache::loadCache(attr(teststat2, "keyList"), dirs = "testStepR"), teststat2) expect_identical(R.cache::loadCache(attr(teststat3, "keyList"), dirs = "testStepR"), teststat3) expect_identical(R.cache::loadCache(attr(teststat4, "keyList"), dirs = "testStepR"), teststat4) expect_identical(R.cache::loadCache(attr(teststat6, "keyList"), dirs = "testStepR"), teststat6) expect_identical(R.cache::loadCache(attr(teststat7, "keyList"), dirs = "testStepR"), teststat7) expect_identical(R.cache::loadCache(attr(teststat8, "keyList"), dirs = "testStepR"), teststat8) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) teststat1 <- monteCarloSimulation(100L, r = 100L, output = "maximum", lengths = c(23:34, 45:67)) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = c(23:34, 45:67), options = list(simulation = "vector", save = list(fileSystem = c("matrixIncreased", "vector", "matrix")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat1, output = "value", lengths = c(23:34, 45:67), options = list())) teststat2 <- monteCarloSimulation(98L, r = 100L, intervalSystem = "dyaPar") teststat3 <- monteCarloSimulation(98L, r = 100L, output = "maximum", intervalSystem = "dyaPar", lengths = 2^(1:3)) expect_identical(critVal(98L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaPar", lengths = 2^(1:3), options = list(simulation = "matrix", save = list(fileSystem = c("matrixIncreased", "vector", "matrix")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(98L, alpha = 0.1, stat = teststat2, output = "value", intervalSystem = "dyaPar", lengths = 2^(1:3), options = list())) teststat <- monteCarloSimulation(127L, r = 100L) teststat4 <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:102, penalty = "log") expect_identical(critVal(102L, alpha = 0.1, r = 100L, output = "value", stat = teststat, penalty = "log", options = list(simulation = "matrixIncreased", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(102L, alpha = 0.1, stat = teststat, output = "value", penalty = "log", options = list())) teststat1 <- monteCarloSimulation(100L, r = 200L, output = "maximum", lengths = c(23:34, 45:67)) expect_identical(critVal(100L, alpha = 0.1, r = 200L, output = "value", lengths = c(23:34, 45:67), options = list(simulation = "vector", save = list(fileSystem = c("matrixIncreased", "vector", "matrix")), load = list(fileSystem = c("vector", "matrix")), dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat1, output = "value", lengths = c(23:34, 45:67), options = list())) expect_identical(critVal(102L, alpha = 0.1, r = 100L, output = "vector", lengths = 1:102, penalty = "log", options = list(simulation = "matrix", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(102L, alpha = 0.1, stat = teststat4, output = "vector", lengths = 1:102, penalty = "log", options = list())) teststat5 <- monteCarloSimulation(73L, r = 100L, output = "maximum", intervalSystem = "dyaLen", lengths = 16L, penalty = "log") expect_identical(critVal(73L, alpha = 0.1, r = 100L, output = "vector", intervalSystem = "dyaLen", lengths = 16L, penalty = "log", stat = teststat5, options = list(simulation = "vector", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(73L, alpha = 0.1, stat = teststat5, output = "vector", intervalSystem = "dyaLen", lengths = 16L, penalty = "log", options = list())) teststat6 <- monteCarloSimulation(125L, r = 100L, intervalSystem = "dyaLen") expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", penalty = "weights", nq = 125L, weights = 1:7 / sum(1:7), intervalSystem = "dyaLen", options = list(simulation = "matrixIncreased", save = list(fileSystem = c("matrixIncreased", "vector")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", penalty = "weights", weights = 1:7 / sum(1:7), intervalSystem = "dyaLen", options = list())) expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", penalty = "log", intervalSystem = "dyaLen", lengths = 2^(4:6), nq = 125L, options = list(simulation = "matrix", save = list(fileSystem = c("matrixIncreased", "vector")), load = list(fileSystem = c("matrixIncreased", "matrix")), dirs = "testStepR")), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", penalty = "log", intervalSystem = "dyaLen", lengths = 2^(4:6), options = list())) teststat <- monteCarloSimulation(74L, r = 200L, output = "maximum", rand.gen = function(data) {rnorm(74)}) expect_identical(critVal(74L, alpha = 0.1, r = 200L, output = "vector", rand.gen = function(data) {rnorm(74)}, options = list(simulation = "vector", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(fileSystem = c("vectorIncreased", "matrix")), dirs = "testStepR")), critVal(74L, alpha = 0.1, stat = teststat, output = "vector", options = list())) teststat7 <- monteCarloSimulation(125L, r = 100L, output = "maximum", intervalSystem = "dyaLen", lengths = 2^(3:5), penalty = "log") expect_identical(critVal(120L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", lengths = 2^(3:5), penalty = "log", nq = 125L, options = list(simulation = "vector", save = list(fileSystem = c("matrixIncreased", "vectorIncreased")), load = list(fileSystem = c("vectorIncreased", "matrixIncreased", "vector")), dirs = "testStepR")), critVal(120L, alpha = 0.1, stat = teststat7, intervalSystem = "dyaLen", lengths = 2^(3:5), penalty = "log", output = "value", options = list())) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 6L) expect_identical(R.cache::loadCache(attr(teststat1, "keyList"), dirs = "testStepR"), teststat1) expect_identical(R.cache::loadCache(attr(teststat2, "keyList"), dirs = "testStepR"), teststat2) expect_identical(R.cache::loadCache(attr(teststat3, "keyList"), dirs = "testStepR"), teststat3) expect_identical(R.cache::loadCache(attr(teststat4, "keyList"), dirs = "testStepR"), teststat4) expect_identical(R.cache::loadCache(attr(teststat6, "keyList"), dirs = "testStepR"), teststat6) expect_identical(R.cache::loadCache(attr(teststat7, "keyList"), dirs = "testStepR"), teststat7) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) }) test_that("stat can be loaded from the stepRdata package", { expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(), load = list(package = "vector")))) expect_error(critVal(100L, alpha = 0.1, r = 100L, options = list(simulation = "vector", save = list(), load = list(package = c(TRUE, FALSE))))) if ("stepRdata" %in% rownames(installed.packages())) { teststat <- monteCarloSimulation(31L) expect_equal(critVal(31L, alpha = 0.1, r = 100L, output = "value", options = list(simulation = "matrixIncreased", load = list(package = TRUE))), critVal(31L, alpha = 0.1, stat = teststat, output = "value")) teststat <- monteCarloSimulation(15L, intervalSystem = "dyaLen") expect_equal(critVal(12L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", options = list(simulation = "matrix", load = list(package = TRUE))), critVal(12L, alpha = 0.1, stat = teststat, output = "value", intervalSystem = "dyaLen")) teststat <- monteCarloSimulation(63L, intervalSystem = "dyaPar") expect_equal(critVal(33L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaPar", options = list(simulation = "vectorIncreased", load = list(package = TRUE))), critVal(33L, alpha = 0.1, stat = teststat, output = "value", intervalSystem = "dyaPar")) } }) test_that("stat can be saved in and loaded from multiple options", { testStepR <- new.env() testvariable <- c("testsavevector", "testsavematrix") testfile <- c(tempfile(pattern = "file", tmpdir = tempdir(), fileext = ".RDS"), tempfile(pattern = "file", tmpdir = tempdir(), fileext = ".RDS")) teststat1 <- monteCarloSimulation(100L, r = 100L, output = "maximum", lengths = c(23:34, 45:67)) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = c(23:34, 45:67), options = list(simulation = "vector", save = list(workspace = c("matrixIncreased", "vector", "matrix"), fileSystem = c("matrixIncreased", "vector", "matrix"), RDSfile = testfile, variable = testvariable), load = list(), envir = testStepR, dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat1, output = "value", lengths = c(23:34, 45:67), options = list())) expect_identical(readRDS(testfile[1]), teststat1) expect_false(file.exists(testfile[2])) expect_identical(get("testsavevector", envir = testStepR), teststat1) expect_false(exists("testsavematrix", envir = testStepR)) teststat2 <- monteCarloSimulation(98L, r = 100L, intervalSystem = "dyaPar") teststat3 <- monteCarloSimulation(98L, r = 100L, output = "maximum", intervalSystem = "dyaPar", lengths = 2^(1:3)) expect_identical(critVal(98L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaPar", lengths = 2^(1:3), options = list(simulation = "matrix", save = list(workspace = c("vector"), fileSystem = c("matrixIncreased", "matrix"), RDSfile = testfile, variable = testvariable), load = list(), envir = testStepR, dirs = "testStepR")), critVal(98L, alpha = 0.1, stat = teststat2, output = "value", intervalSystem = "dyaPar", lengths = 2^(1:3), options = list())) expect_identical(readRDS(testfile[1]), teststat3) expect_identical(readRDS(testfile[2]), teststat2) expect_identical(get("testsavevector", envir = testStepR), teststat3) expect_identical(get("testsavematrix", envir = testStepR), teststat2) unlink(testfile) remove(testsavevector, envir = testStepR) teststat <- monteCarloSimulation(127L, r = 100L) teststat4 <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 1:102, penalty = "log") expect_identical(critVal(102L, alpha = 0.1, r = 100L, output = "value", stat = teststat, penalty = "log", options = list(simulation = "matrixIncreased", save = list(workspace = c("matrix", "vectorIncreased"), fileSystem = c("vector", "matrix")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR, dirs = "testStepR")), critVal(102L, alpha = 0.1, stat = teststat, output = "value", penalty = "log", options = list())) teststat1new <- monteCarloSimulation(100L, r = 200L, output = "maximum", lengths = c(23:34, 45:67)) expect_identical(critVal(100L, alpha = 0.1, r = 200L, output = "value", lengths = c(23:34, 45:67), options = list(simulation = "vector", save = list(workspace = c("matrixIncreased", "vector", "matrix")), load = list(workspace = c("vector", "matrix")), envir = testStepR, dirs = "testStepR")), critVal(100L, alpha = 0.1, stat = teststat1new, output = "value", lengths = c(23:34, 45:67), options = list())) expect_identical(critVal(102L, alpha = 0.1, r = 50L, output = "vector", lengths = 1:102, penalty = "log", options = list(simulation = "matrix", save = list(fileSystem = c("matrix", "vectorIncreased")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR, dirs = "testStepR")), critVal(102L, alpha = 0.1, stat = teststat4, output = "vector", lengths = 1:102, penalty = "log", options = list())) teststat5 <- monteCarloSimulation(73L, r = 100L, output = "maximum", intervalSystem = "dyaLen", lengths = 16L, penalty = "log") expect_identical(critVal(73L, alpha = 0.1, r = 100L, output = "vector", intervalSystem = "dyaLen", lengths = 16L, penalty = "log", stat = teststat5, options = list(simulation = "vector", save = list(fileSystem = c("matrix", "vectorIncreased", "vector"), workspace = c("matrix", "matrixIncreased")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR, dirs = "testStepR")), critVal(73L, alpha = 0.1, stat = teststat5, output = "vector", intervalSystem = "dyaLen", lengths = 16L, penalty = "log", options = list())) teststat6 <- monteCarloSimulation(125L, r = 100L, intervalSystem = "dyaLen") expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", penalty = "weights", nq = 125L, weights = 1:7 / sum(1:7), intervalSystem = "dyaLen", options = list(simulation = "matrixIncreased", save = list(workspace = c("matrixIncreased", "vector"), fileSystem = c("matrix", "vectorIncreased")), load = list(workspace = c("vectorIncreased", "matrix")), envir = testStepR, dirs = "testStepR")), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", penalty = "weights", weights = 1:7 / sum(1:7), intervalSystem = "dyaLen", options = list())) teststat6b <- monteCarloSimulation(125L, r = 100L, output = "maximum", penalty = "log", intervalSystem = "dyaLen", lengths = 2^(4:6)) expect_identical(critVal(107L, alpha = 0.1, r = 100L, output = "vector", penalty = "log", intervalSystem = "dyaLen", lengths = 2^(4:6), nq = 125L, options = list(simulation = "matrix", save = list(workspace = c("matrixIncreased", "vector"), fileSystem = c("matrix", "vectorIncreased")), load = list(workspace = c("matrixIncreased", "matrix")), envir = testStepR, dirs = "testStepR")), critVal(107L, alpha = 0.1, stat = teststat6, output = "vector", penalty = "log", intervalSystem = "dyaLen", lengths = 2^(4:6), options = list())) teststat <- monteCarloSimulation(74L, r = 200L, output = "maximum", rand.gen = function(data) {rnorm(74)}) expect_identical(critVal(74L, alpha = 0.1, r = 200L, output = "vector", rand.gen = function(data) {rnorm(74)}, options = list(simulation = "vector", save = list(workspace = c("matrix", "vector")), load = list(fileSystem = c("vectorIncreased", "matrix")), envir = testStepR, dirs = "testStepR")), critVal(74L, alpha = 0.1, stat = teststat, output = "vector", options = list())) teststat7 <- monteCarloSimulation(125L, r = 100L, output = "maximum", intervalSystem = "dyaLen", lengths = 2^(3:5), penalty = "log") expect_identical(critVal(120L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", lengths = 2^(3:5), penalty = "log", nq = 125L, options = list(simulation = "vectorIncreased", save = list(workspace = c("matrixIncreased", "vectorIncreased")), load = list(workspace = c("matrix", "vector"), fileSystem = c("vectorIncreased", "matrixIncreased")), envir = testStepR, dirs = "testStepR")), critVal(120L, alpha = 0.1, stat = teststat7, output = "value", options = list(), intervalSystem = "dyaLen", lengths = 2^(3:5), penalty = "log")) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 5L) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat1new) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[2]], teststat3) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[3]], teststat4) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[4]], teststat6) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[5]], teststat7) remove(critValStepRTab, envir = testStepR) expect_identical(length(list.files(file.path(R.cache::getCacheRootPath(), "testStepR"))), 5L) expect_identical(R.cache::loadCache(attr(teststat1, "keyList"), dirs = "testStepR"), teststat1) expect_identical(R.cache::loadCache(attr(teststat2, "keyList"), dirs = "testStepR"), teststat2) expect_identical(R.cache::loadCache(attr(teststat4, "keyList"), dirs = "testStepR"), teststat4) expect_identical(R.cache::loadCache(attr(teststat5, "keyList"), dirs = "testStepR"), teststat5) expect_identical(R.cache::loadCache(attr(teststat6b, "keyList"), dirs = "testStepR"), teststat6b) unlink(file.path(R.cache::getCacheRootPath(), "testStepR"), recursive = TRUE) }) test_that("... is tested and works", { expect_error(critVal(100L, alpha = 0.1, test = 1, options = list(load = list()))) expect_error(critVal(100L, alpha = 0.1, sd = "s", options = list(load = list()))) expect_error(critVal(100L, alpha = 0.1, sd = 0, options = list(load = list()))) expect_error(critVal(100L, alpha = 0.1, sd = c(1, 2), options = list(load = list()))) teststat <- monteCarloSimulation(n = 15L) expect_identical(critVal(12, alpha = 0.1, sd = 5, options = list(load = list())), critVal(12, alpha = 0.1, stat = teststat)) expect_error(supressWarning(critVal(100L, alpha = 0.1, r = 100L, seed = "s", options = list(load = list())))) expect_identical(critVal(12, alpha = 0.1, r = 100L, seed = c(1, 2), options = list(load = list())), critVal(12, alpha = 0.1, r = 100L, seed = 1L, options = list(load = list()))) expect_identical(critVal(12, alpha = 0.1, r = 100L, seed = 100.5, options = list(load = list())), critVal(12, alpha = 0.1, r = 100L, seed = 100L, options = list(load = list()))) teststat <- monteCarloSimulation(n = 15L, r = 100L) expect_identical(critVal(12, alpha = 0.1, r = 100L, seed = 15L, options = list(load = list())), critVal(12, alpha = 0.1, r = 100L, stat = teststat)) expect_error(critVal(10L, alpha = 0.1, r = 100L, rand.gen = 10, options = list(load = list()))) expect_error(critVal(10L, alpha = 0.1, r = 100L, rand.gen = function(data, n) {rnorm(10)}, options = list(load = list()))) expect_error(critVal(10L, alpha = 0.1, r = 100L, rand.gen = function(data) {rnorm(10)}, options = list(load = list()))) teststat <- monteCarloSimulation(n = 15L, r = 100L) expect_identical(critVal(12, alpha = 0.1, r = 100L, rand.gen = function(data) {rnorm(15)}, options = list(load = list())), critVal(12, alpha = 0.1, r = 100L, stat = teststat)) expect_error(critVal(10L, alpha = 0.1, r = 100L, messages = "s", options = list(load = list()))) expect_error(critVal(10L, alpha = 0.1, r = 100L, messages = c(10, 20), options = list(load = list()))) expect_error(critVal(10L, alpha = 0.1, r = 100L, messages = 0, options = list(load = list()))) expect_identical(suppressMessages(critVal(12, alpha = 0.1, r = 100L, messages = 10.5, options = list(load = list()))), suppressMessages(critVal(12, alpha = 0.1, r = 100L, messages = 10L, options = list(load = list())))) }) test_that("family 'hsmuce' works", { testn <- 5L teststat <- monteCarloSimulation(n = 32L, r = 100L, family = "hsmuce") expect_error(critVal(stat = teststat, family = "hsmuce")) expect_warning(ret <- critVal(n = 2^testn, stat = teststat, family = "hsmuce")) expect_identical(ret, critVal(alpha = 0.5, n = 2^testn, stat = teststat, nq = 2^testn, family = "hsmuce", intervalSystem = "dyaPar", lengths = 2^(1:5), penalty = "weights", output = "vector")) testVector(teststat = teststat, testn = as.integer(2^testn), testnq = as.integer(2^testn), testfamily = "hsmuce", testlengths = 2^(1:testn), testalpha = 0.1, testweights = rep(0.2, 5), defaultPenalty = "weights") testValue(teststat = teststat, testn = as.integer(2^testn), testnq =as.integer(2^testn), testfamily = "hsmuce", testlengths = 2^(1:testn), testalpha = 0.1, testweights = rep(0.2, 5), defaultPenalty = "weights") expect_error(critVal(q = "s", alpha = 0.1, n = as.integer(2^testn), stat = teststat, family = "hsmuce")) expect_error(critVal(q = rep(1, 8), alpha = 0.1, n = as.integer(2^testn), stat = teststat, family = "hsmuce")) expect_error(critVal(q = rep(1, 4), alpha = 0.1, n = as.integer(2^testn), stat = teststat, family = "hsmuce")) testcompare <- 1 attr(testcompare, "n") <- as.integer(2^testn) expect_identical(critVal(q = 1, n = as.integer(2^testn), stat = teststat, output = "value", family = "hsmuce", penalty = "sqrt"), testcompare) testcompare <- rep(1, 5) attr(testcompare, "n") <- as.integer(2^testn) expect_identical(critVal(q = rep(1, 5), n = as.integer(2^testn), stat = teststat, family = "hsmuce"), testcompare) testn <- 10L teststat <- monteCarloSimulation(n = 1024L, r = 100L, family = "hsmuce") expect_error(critVal(alpha = 0, n = 2^testn, stat = teststat, family = "hsmuce")) testDetailedHSMUCE(teststat = teststat, testn = as.integer(2^testn), testnq = 2^testn, testfamily = "hsmuce", testlengths = 2^(1:testn), testalpha = 0.0567, testweights = rep(0.1, 10), defaultPenalty = "weights") # testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^testn, testfamily = "hsmuce", # testlengths = 2^(1:testn), testalpha = 0.0567, testweights = rep(0.1, 10), # defaultPenalty = "weights") testValue(teststat = teststat, testn = 2^testn, testnq = 2^testn, testfamily = "hsmuce", testlengths = 2^(1:testn), testalpha = 0.0567, testweights = rep(0.1, 10), defaultPenalty = "weights") expect_error(critVal(n = "s", alpha = 0.1, stat = teststat, family = "hsmuce")) expect_error(critVal(n = c(1L, 2L), alpha = 0.1, stat = teststat, family = "hsmuce")) expect_error(critVal(n = Inf, alpha = 0.1, stat = teststat, family = "hsmuce")) expect_identical(critVal(n = 1010, alpha = 0.1, stat = teststat, family = "hsmuce"), critVal(n = 1010L, alpha = 0.1, stat = teststat, family = "hsmuce")) expect_identical(critVal(n = 1010.5, alpha = 0.1, stat = teststat, family = "hsmuce"), critVal(n = 1010L, alpha = 0.1, stat = teststat, family = "hsmuce")) expect_identical(critVal(n = 2^testn, alpha = 0.1, stat = teststat, family = "hsmuce"), critVal(n = 2^testn, nq = 2^testn, alpha = 0.1, stat = teststat, family = "hsmuce")) testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^20L, testfamily = "hsmuce", testlengths = 2^(1:testn), testalpha = 0.01, testweights = rep(0.1, 10), defaultPenalty = "weights") testValue(teststat = teststat, testn = 2^testn, testnq = 2^20L, testfamily = "hsmuce", testlengths = 2^(1:testn), testalpha = 0.01, testweights = rep(0.1, 10), defaultPenalty = "weights") expect_error(critVal(n = 2^testn, penalty = "", alpha = 0.1, stat = teststat, family = "hsmuce")) expect_error(critVal(n = 2^testn, penalty = c("gauss"), alpha = 0.1, stat = teststat, family = "hsmuce")) expect_identical(critVal(n = 2^testn, alpha = 0.1, stat = teststat, family = "hsmuce"), critVal(n = 2^testn, penalty = "weights", alpha = 0.1, stat = teststat, family = "hsmuce")) testn <- 5L teststat <- monteCarloSimulation(n = 32L, r = 100L, family = "hsmuce") expect_identical(critVal(n = 2^testn, alpha = 0.1, penalty = "weights", stat = teststat, family = "hsmuce"), critVal(n = 2^testn, alpha = 0.1, penalty = "weights", weights = rep(1 / testn, testn), stat = teststat, family = "hsmuce")) expect_identical(critVal(n = 2^testn, penalty = "weights", weights = rep(1, testn), alpha = 0.1, stat = teststat, family = "hsmuce"), critVal(n = 2^testn, penalty = "weights", weights = rep(1 / testn, testn), alpha = 0.1, stat = teststat, family = "hsmuce")) testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^20L, testlengths = 2^(1:testn), testalpha = 0.01, compareWeights = rep(1 / testn, testn), defaultPenalty = "weights", testfamily = "hsmuce") testValue(teststat = teststat, testn = 2^testn, testnq = 2^20L, testlengths = 2^(1:testn), testalpha = 0.01, defaultPenalty = "weights", testfamily = "hsmuce") testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^20L, testlengths = 2^(1:testn), testalpha = 0.01, testweights = 5:1 / sum(5:1), defaultPenalty = "weights", testfamily = "hsmuce") testValue(teststat = teststat, testn = 2^testn, testnq = 2^20L, testlengths = 2^(1:testn), testalpha = 0.01, testweights = 5:1 / sum(5:1), defaultPenalty = "weights", testfamily = "hsmuce") testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^testn, testlengths = 2^(1:testn), testalpha = 0.01, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "weights", testfamily = "hsmuce") testValue(teststat = teststat, testn = as.integer(2^testn), testnq = 2^testn, testlengths = 2^(1:testn), testalpha = 0.01, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "weights", testfamily = "hsmuce") testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^testn, testlengths = 2^(1:testn), testalpha = 1e-6, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "weights", testfamily = "hsmuce") testValue(teststat = teststat, testn = 2^testn, testnq = 2^testn, testlengths = 2^(1:testn), testalpha = 1e-6, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "weights", testfamily = "hsmuce") testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^testn, testlengths = 2^(1:testn), testalpha = 1 - 1e-6, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "weights", testfamily = "hsmuce") testValue(teststat = teststat, testn = 2^testn, testnq = 2^testn, testlengths = 2^(1:testn), testalpha = 1- 1e-6, testweights = c(5, 3, 6, 3, 9) / sum(c(5, 3, 6, 3, 9)), defaultPenalty = "weights", testfamily = "hsmuce") expect_identical(critVal(n = 2^testn, alpha = 0.1, stat = teststat, family = "hsmuce"), critVal(n = 2^testn, alpha = 0.1, stat = teststat, family = "hsmuce", intervalSystem = "dyaPar")) teststat <- monteCarloSimulation(n = 6L, r = 100L, family = "hsmuce", intervalSystem = "all") testVector(teststat = teststat, testn = 6L, testnq = 20L, testlengths = 1:testn + 1, testalpha = 0.1, defaultPenalty = "weights", testweights = rep(1 / testn, testn), testintervalsystem = "all", testfamily = "hsmuce") testValue(teststat = teststat, testn = 6L, testnq = 20L, testlengths = 1:testn + 1, testalpha = 0.1, defaultPenalty = "weights", testintervalsystem = "all", testfamily = "hsmuce") teststat <- monteCarloSimulation(n = 32L, r = 100L, family = "hsmuce", intervalSystem = "dyaLen") testVector(teststat = teststat, testn = 32L, testnq = 32L, testlengths = 2^(1:5), testalpha = 0.1, defaultPenalty = "weights", testweights = rep(1 / testn, testn), testintervalsystem = "dyaLen", testfamily = "hsmuce") testValue(teststat = teststat, testn = 32L, testnq = 32L, testlengths = 2^(1:5), testalpha = 0.1, defaultPenalty = "weights", testintervalsystem = "dyaLen", testfamily = "hsmuce") testn <- 6L teststat <- monteCarloSimulation(n = 32L, r = 100L, family = "hsmuce") expect_error(critVal(n = testn, alpha = 0.1, stat = teststat, lengths = "s", family = "hsmuce")) expect_error(critVal(n = 2^testn, alpha = 0.1, stat = teststat, lengths = 2^(0:5), family = "hsmuce")) expect_error(critVal(n = 2^testn, alpha = 0.1, stat = teststat, lengths = c(2^(2:6), 0L), family = "hsmuce")) expect_error(critVal(q = 1, n = testn, stat = teststat, output = "value", family = "hsmuce")) compare <- 1 attr(compare, "n") <- testn expect_identical(critVal(q = 1, n = testn, stat = teststat, output = "value", family = "hsmuce", penalty = "sqrt"), compare) compare <- rep(1, 5) attr(compare, "n") <- 32L expect_identical(critVal(q = rep(1, 5), n = 32L, stat = teststat, family = "hsmuce"), compare) testq <- 1 attr(testq, "n") <- 10L compare <- 1 attr(compare, "n") <- 10L expect_identical(critVal(q = testq, n = testn, stat = teststat, output = "value", family = "hsmuce", penalty = "log"), compare) testq <- rep(1, 5) attr(testq, "n") <- 100L compare <- rep(1, 5) attr(compare, "n") <- 100L expect_identical(critVal(q = testq, n = 32L, stat = teststat, family = "hsmuce"), compare) testq <- as.numeric(2^(1:5)) attr(testq, "n") <- 32L compare <- as.numeric(2^(1:5)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaLen", family = "hsmuce"), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 32L compare <- as.numeric(2^(1:5)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaLen", family = "hsmuce"), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 64L compare <- as.numeric(2^(1:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaPar", family = "hsmuce"), compare) testq <- as.numeric(1:64) attr(testq, "n") <- 64L compare <- as.numeric(2^(1:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaPar", family = "hsmuce"), compare) testq <- as.numeric(2^(1:6)) attr(testq, "n") <- 64L compare <- as.numeric(2^(1:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "dyaPar", family = "hsmuce"), compare) testq <- as.numeric(2:32) attr(testq, "n") <- 32L compare <- as.numeric(2:32) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "all", family = "hsmuce"), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 32L compare <- as.numeric(2:32) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "all", family = "hsmuce"), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 64L compare <- as.numeric(2:32) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "all", family = "hsmuce"), compare) testq <- as.numeric(1:64) attr(testq, "n") <- 64L compare <- as.numeric(2:32) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "all", family = "hsmuce"), compare) testq <- as.numeric(2:64) attr(testq, "n") <- 64L compare <- as.numeric(2:32) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, intervalSystem = "all", family = "hsmuce"), compare) testq <- as.numeric(1:5) compare <- as.numeric(1:5) attr(compare, "n") <- 256L expect_identical(critVal(q = testq, n = 256L, lengths = 2^(1:5), family = "hsmuce"), compare) testq <- as.numeric(1:8) compare <- as.numeric(1:5) attr(compare, "n") <- 256L expect_identical(critVal(q = testq, n = 256L, lengths = 2^(1:5), family = "hsmuce"), compare) testq <- as.numeric(3:7) attr(testq, "n") <- 4096L compare <- as.numeric(3:7) attr(compare, "n") <- 4096L expect_identical(critVal(q = testq, n = 256L, lengths = 2^(3:7), family = "hsmuce"), compare) testq <- as.numeric(1:8) attr(testq, "n") <- 4096L compare <- as.numeric(3:7) attr(compare, "n") <- 4096L expect_identical(critVal(q = testq, n = 256L, lengths = 2^(3:7), family = "hsmuce"), compare) testq <- as.numeric(1:12) attr(testq, "n") <- 4096L compare <- as.numeric(3:7) attr(compare, "n") <- 4096L expect_identical(critVal(q = testq, n = 256L, lengths = 2^(3:7), family = "hsmuce"), compare) testq <- as.numeric(1:4096) attr(testq, "n") <- 4096L compare <- as.numeric(2^(3:7)) attr(compare, "n") <- 4096L expect_identical(critVal(q = testq, n = 256L, lengths = 2^(3:7), family = "hsmuce"), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 32L compare <- as.numeric(2^(1:4)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 16L, lengths = 2^(1:4), intervalSystem = "all", family = "hsmuce"), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 32L compare <- as.numeric(2^(2:3)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 16L, lengths = 2^(2:3), intervalSystem = "all", family = "hsmuce"), compare) testq <- as.numeric(1:16) attr(testq, "n") <- 32L compare <- as.numeric(2^(2:3)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 16L, lengths = 2^(2:3), intervalSystem = "all", family = "hsmuce"), compare) testq <- as.numeric(2^(2:3)) attr(testq, "n") <- 32L compare <- as.numeric(2^(2:3)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 16L, lengths = 2^(2:3), intervalSystem = "all", family = "hsmuce"), compare) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum", intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log", family = "hsmuce") expect_identical(critVal(100L, alpha = 0.1, r = 100L, family = "hsmuce", lengths = c(2, 4, 8), penalty = "log", options = list(simulation = "vector", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, family = "hsmuce", lengths = c(2, 4, 8), penalty = "log")) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 2:30, family = "hsmuce", intervalSystem = "all", penalty = "sqrt") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = 2:30, family = "hsmuce", intervalSystem = "all", penalty = "sqrt", options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = 2:30, family = "hsmuce", intervalSystem = "all", penalty = "sqrt")) teststat <- monteCarloSimulation(120L, r = 100L, output = "maximum", lengths = 2:30, family = "hsmuce", intervalSystem = "all", penalty = "sqrt") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = 2:30, family = "hsmuce", intervalSystem = "all", penalty = "sqrt", nq = 120L, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = 2:30, family = "hsmuce", intervalSystem = "all", penalty = "sqrt")) teststat <- monteCarloSimulation(100L, r = 100L, family = "hsmuce") expect_identical(critVal(100L, alpha = 0.1, r = 100L, lengths = c(2, 4, 32), family = "hsmuce", options = list(simulation = "matrix", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, lengths = c(2, 4, 32), family = "hsmuce")) teststat <- monteCarloSimulation(127L, r = 100L, intervalSystem = "dyaLen", family = "hsmuce") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 32, 64), family = "hsmuce", options = list(simulation = "matrixIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 32, 64), family = "hsmuce")) testStepR <- new.env() teststat <- monteCarloSimulation(100L, r = 100L, family = "hsmuce") expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "vector", family = "hsmuce", options = list(simulation = "matrix", save = list(workspace = "matrix"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", family = "hsmuce", options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "vector", family = "hsmuce", options = list(simulation = "matrix", save = list(), load = list(workspace = "matrix"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", options = list(), family = "hsmuce")) remove(critValStepRTab, envir = testStepR) if ("stepRdata" %in% rownames(installed.packages())) { teststat <- monteCarloSimulation(31L, family = "hsmuce", intervalSystem = "all") expect_equal(critVal(31L, alpha = 0.1, r = 100L, family = "hsmuce", intervalSystem = "all", options = list(simulation = "matrixIncreased", load = list(package = TRUE))), critVal(31L, alpha = 0.1, stat = teststat, family = "hsmuce", intervalSystem = "all")) teststat <- monteCarloSimulation(15L, family = "hsmuce", intervalSystem = "dyaLen") expect_equal(critVal(12L, alpha = 0.1, r = 100L, family = "hsmuce", intervalSystem = "dyaLen", options = list(simulation = "matrixIncreased", load = list(package = TRUE))), critVal(12L, alpha = 0.1, stat = teststat, family = "hsmuce", intervalSystem = "dyaLen")) teststat <- monteCarloSimulation(63L, family = "hsmuce", intervalSystem = "dyaPar") expect_equal(critVal(33L, alpha = 0.1, r = 100L, family = "hsmuce", intervalSystem = "dyaPar", options = list(simulation = "matrixIncreased", load = list(package = TRUE))), critVal(33L, alpha = 0.1, stat = teststat, family = "hsmuce", intervalSystem = "dyaPar")) } }) test_that("family 'mDependentPS' works", { testthat::skip_on_cran() testCovariances <- seq(1, 0.2, -0.2) teststat <- monteCarloSimulation(n = 16L, r = 100L, family = "mDependentPS", covariances = testCovariances) expect_error(critVal(stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_warning(ret <- critVal(n = 16L, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_identical(ret, critVal(alpha = 0.5, n = 16L, stat = teststat, nq = 16L, family = "mDependentPS", covariances = testCovariances, intervalSystem = "dyaLen", lengths = 2^(0:4), penalty = "sqrt", output = "vector")) testVector(teststat = teststat, testn = 16L, testnq = 16L, testfamily = "mDependentPS", testlengths = 2^(0:4), testalpha = 0.1, testweights = rep(0.2, 5), defaultPenalty = "sqrt", covariances = testCovariances) testValue(teststat = teststat, testn = 16L, testnq = 16L, testfamily = "mDependentPS", testlengths = 2^(0:4), testalpha = 0.1, testweights = rep(0.2, 5), defaultPenalty = "sqrt", covariances = testCovariances) expect_error(critVal(q = "s", alpha = 0.1, n = 16L, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_error(critVal(q = rep(1, 8), alpha = 0.1, n = 16L, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_error(critVal(q = rep(1, 4), alpha = 0.1, n = 16L, stat = teststat, family = "mDependentPS", covariances = testCovariances)) testn <- 10L testCovariances <- c(1, 0.56, 0.234, 0.123) teststat <- monteCarloSimulation(n = 1024L, r = 100L, family = "mDependentPS", covariances = testCovariances) expect_error(critVal(alpha = 0, n = 2^testn, stat = teststat, family = "mDependentPS", covariances = testCovariances)) testVector(teststat = teststat, testn = 1024L, testnq = 2^testn, testfamily = "mDependentPS", testlengths = 2^(0:testn), testalpha = 0.0567, testweights = rep(1 / 11, 11), defaultPenalty = "sqrt", covariances = testCovariances) testValue(teststat = teststat, testn = 2^testn, testnq = 2^testn, testfamily = "mDependentPS", testlengths = 2^(0:testn), testalpha = 0.0567, testweights = rep(1 / 11, 11), defaultPenalty = "sqrt", covariances = testCovariances) expect_error(critVal(n = "s", alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_error(critVal(n = c(1L, 2L), alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_error(critVal(n = Inf, alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_error(critVal(n = 0L, alpha = 0.1, stat = as.matrix(teststat[1, ]), family = "mDependentPS", covariances = testCovariances)) expect_identical(critVal(n = 1010, alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances), critVal(n = 1010L, alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_identical(critVal(n = 1010.5, alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances), critVal(n = 1010L, alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_identical(critVal(n = 2^testn, alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances), critVal(n = 2^testn, nq = 2^testn, alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) testVector(teststat = teststat, testn = 1024L, testnq = 2^20L, testfamily = "mDependentPS", testlengths = 2^(0:testn), testalpha = 0.01, testweights = rep(1 / 11, 11), defaultPenalty = "sqrt", covariances = testCovariances) testValue(teststat = teststat, testn = 2^testn, testnq = 2^20L, testfamily = "mDependentPS", testlengths = 2^(0:testn), testalpha = 0.01, testweights = rep(1 / 11, 11), defaultPenalty = "sqrt", covariances = testCovariances) expect_error(critVal(n = 2^testn, penalty = "", alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_error(critVal(n = 2^testn, penalty = c("gauss"), alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_identical(critVal(n = 2^testn, alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances), critVal(n = 2^testn, penalty = "sqrt", alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) testn <- 5L testCovariances <- c(3, 2, 0.5, -1) teststat <- monteCarloSimulation(n = 32L, r = 100L, family = "mDependentPS", covariances = testCovariances) expect_identical(critVal(n = 2^testn, alpha = 0.1, penalty = "weights", stat = teststat, family = "mDependentPS", covariances = testCovariances), critVal(n = 2^testn, alpha = 0.1, penalty = "weights", weights = rep(1 / 6, 6), stat = teststat, family = "mDependentPS", covariances = testCovariances)) expect_identical(critVal(n = 2^testn, penalty = "weights", weights = rep(1, 6), alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances), critVal(n = 2^testn, penalty = "weights", weights = rep(1 / 6, 6), alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances)) testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^20L, testlengths = 2^(0:testn), testalpha = 0.01, compareWeights = rep(1 / 6, 6), defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) testValue(teststat = teststat, testn = 2^testn, testnq = 2^20L, testlengths = 2^(0:testn), testalpha = 0.01, defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^20L, testlengths = 2^(0:testn), testalpha = 0.01, testweights = 6:1 / sum(6:1), defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) testValue(teststat = teststat, testn = 2^testn, testnq = 2^20L, testlengths = 2^(0:testn), testalpha = 0.01, testweights = 6:1 / sum(6:1), defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^testn, testlengths = 2^(0:testn), testalpha = 0.01, testweights = c(5, 3, 6, 3, 9, 1) / sum(c(5, 3, 6, 3, 9, 1)), defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) testValue(teststat = teststat, testn = 2^testn, testnq = 2^testn, testlengths = 2^(0:testn), testalpha = 0.01, testweights = c(5, 3, 6, 3, 9, 1) / sum(c(5, 3, 6, 3, 9, 1)), defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^testn, testlengths = 2^(0:testn), testalpha = 1e-6, testweights = c(5, 3, 6, 3, 9, 6) / sum(c(5, 3, 6, 3, 9, 6)), defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) testValue(teststat = teststat, testn = 2^testn, testnq = 2^testn, testlengths = 2^(0:testn), testalpha = 1e-6, testweights = c(5, 3, 6, 3, 9, 6) / sum(c(5, 3, 6, 3, 9, 6)), defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) testVector(teststat = teststat, testn = as.integer(2^testn), testnq = 2^testn, testlengths = 2^(0:testn), testalpha = 1 - 1e-6, testweights = c(5, 3, 6, 3, 9, 13) / sum(c(5, 3, 6, 3, 9, 13)), defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) testValue(teststat = teststat, testn = 2^testn, testnq = 2^testn, testlengths = 2^(0:testn), testalpha = 1- 1e-6, testweights = c(5, 3, 6, 3, 9, 13) / sum(c(5, 3, 6, 3, 9, 13)), defaultPenalty = "sqrt", testfamily = "mDependentPS", covariances = testCovariances) expect_identical(critVal(n = as.integer(2^testn), alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances), critVal(n = 2^testn, alpha = 0.1, stat = teststat, family = "mDependentPS", covariances = testCovariances, intervalSystem = "dyaLen")) teststat <- monteCarloSimulation(n = 12L, r = 100L, family = "mDependentPS", covariances = testCovariances, intervalSystem = "all") testVector(teststat = teststat, testn = 12L, testnq = 20L, testlengths = 1:6, compareStat = teststat[1:6, ], testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "all", testfamily = "mDependentPS", covariances = testCovariances) testValue(teststat = teststat, testn = 12L, testnq = 20L, testlengths = 1:6, testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "all", testfamily = "mDependentPS", covariances = testCovariances) teststat <- monteCarloSimulation(n = 32L, r = 100L, family = "mDependentPS", covariances = testCovariances, intervalSystem = "dyaPar") testVector(teststat = teststat, testn = 32L, testnq = 32L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 6, 6), testintervalsystem = "dyaPar", testfamily = "mDependentPS", covariances = testCovariances) testValue(teststat = teststat, testn = 32L, testnq = 32L, testlengths = 2^(0:5), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaPar", testfamily = "mDependentPS", covariances = testCovariances) compare <- 1 attr(compare, "n") <- testn expect_identical(critVal(q = 1, n = testn, stat = teststat, output = "value", family = "mDependentPS", covariances = testCovariances), compare) compare <- rep(1, 5) attr(compare, "n") <- 16L expect_identical(critVal(q = rep(1, 5), n = 16L, stat = teststat, family = "mDependentPS", covariances = testCovariances), compare) testq <- 1 attr(testq, "n") <- 10L compare <- 1 attr(compare, "n") <- 10L expect_identical(critVal(q = testq, n = testn, stat = teststat, output = "value", family = "mDependentPS", covariances = testCovariances), compare) testq <- rep(1, 5) attr(testq, "n") <- 20L compare <- rep(1, 5) attr(compare, "n") <- 20L expect_identical(critVal(q = testq, n = 16L, stat = teststat, family = "mDependentPS", covariances = testCovariances), compare) testq <- as.numeric(2^(0:5)) attr(testq, "n") <- 32L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 32L, family = "mDependentPS", covariances = testCovariances), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 32L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 32L expect_identical(critVal(q = testq, n = 32L, family = "mDependentPS", covariances = testCovariances), compare) testq <- as.numeric(1:32) attr(testq, "n") <- 64L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, family = "mDependentPS", covariances = testCovariances), compare) testq <- as.numeric(1:64) attr(testq, "n") <- 64L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, family = "mDependentPS", covariances = testCovariances), compare) testq <- as.numeric(2^(0:6)) attr(testq, "n") <- 64L compare <- as.numeric(2^(0:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, family = "mDependentPS", covariances = testCovariances), compare) teststat <- monteCarloSimulation(100L, r = 100L, output = "maximum", intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log", family = "mDependentPS", covariances = testCovariances) expect_identical(critVal(100L, alpha = 0.1, r = 100L, intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log", family = "mDependentPS", covariances = testCovariances, options = list(simulation = "vector", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, intervalSystem = "dyaPar", lengths = c(2, 4, 8), penalty = "log", family = "mDependentPS", covariances = testCovariances)) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 2:30) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = 2:30, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = 2:30)) teststat <- monteCarloSimulation(120L, r = 100L, output = "maximum", lengths = c(2, 4, 8), family = "mDependentPS", covariances = testCovariances) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = c(2, 4, 8), nq = 120L, family = "mDependentPS", covariances = testCovariances, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = c(2, 4, 8), family = "mDependentPS", covariances = testCovariances)) teststat <- monteCarloSimulation(100L, r = 100L, family = "mDependentPS", covariances = testCovariances) expect_identical(critVal(100L, alpha = 0.1, r = 100L, lengths = c(2, 8, 64), family = "mDependentPS", covariances = testCovariances, options = list(simulation = "matrix", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, lengths = c(2, 8, 64), family = "mDependentPS", covariances = testCovariances)) teststat <- monteCarloSimulation(127L, r = 100L, intervalSystem = "dyaLen", family = "mDependentPS", covariances = testCovariances) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 32, 64), family = "mDependentPS", covariances = testCovariances, options = list(simulation = "matrixIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 32, 64), family = "mDependentPS", covariances = testCovariances)) testStepR <- new.env() teststat <- monteCarloSimulation(100L, r = 100L, family = "mDependentPS", covariances = testCovariances) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "vector", family = "mDependentPS", covariances = testCovariances, options = list(simulation = "matrix", save = list(workspace = "matrix"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", family = "mDependentPS", covariances = testCovariances, options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "vector", family = "mDependentPS", covariances = testCovariances + 1e-12, options = list(simulation = "matrix", save = list(), load = list(workspace = "matrix"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", options = list(), family = "mDependentPS", covariances = testCovariances)) remove(critValStepRTab, envir = testStepR) }) test_that("family 'jsmurf' works", { testthat::skip_on_cran() testfilter <- lowpassFilter::lowpassFilter(type = "bessel", param = list(pole = 4L, cutoff = 0.1), sr = 1e4) teststat <- monteCarloSimulation(n = 64L, r = 100L, family = "jsmurf", filter = testfilter) expect_error(critVal(stat = teststat, family = "jsmurf", filter = testfilter)) expect_warning(ret <- critVal(n = 64L, stat = teststat, family = "jsmurf", filter = testfilter)) expect_identical(ret, critVal(alpha = 0.5, n = 64L, stat = teststat, nq = 64L, family = "jsmurf", filter = testfilter, intervalSystem = "dyaLen", lengths = 2^(4:6), penalty = "sqrt", output = "vector")) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "jsmurf", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "jsmurf", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter) teststat <- monteCarloSimulation(n = 64L, r = 100L, family = "jsmurfPS", filter = testfilter) expect_error(critVal(stat = teststat, family = "jsmurfPS", filter = testfilter)) expect_warning(ret <- critVal(n = 64L, stat = teststat, family = "jsmurfPS", filter = testfilter)) expect_identical(ret, critVal(alpha = 0.5, n = 64L, stat = teststat, nq = 64L, family = "jsmurfPS", filter = testfilter, intervalSystem = "dyaLen", lengths = 2^(4:6), penalty = "sqrt", output = "vector")) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "jsmurfPS", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "jsmurfPS", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter) teststat <- monteCarloSimulation(n = 64L, r = 100L, family = "jsmurfLR", filter = testfilter) expect_error(critVal(stat = teststat, family = "jsmurfLR", filter = testfilter)) expect_warning(ret <- critVal(n = 64L, stat = teststat, family = "jsmurfLR", filter = testfilter)) expect_identical(ret, critVal(alpha = 0.5, n = 64L, stat = teststat, nq = 64L, family = "jsmurfLR", filter = testfilter, intervalSystem = "dyaLen", lengths = 2^(4:6), penalty = "sqrt", output = "vector")) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "jsmurfLR", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "jsmurfLR", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter) expect_error(critVal(q = "s", alpha = 0.1, n = 16L, stat = teststat, family = "jsmurf", filter = testfilter)) expect_error(critVal(q = rep(1, 8), alpha = 0.1, n = 32L, family = "jsmurf", filter = testfilter)) expect_error(critVal(alpha = 0, n = 32L, family = "jsmurf", filter = testfilter)) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "jsmurfLR", testlengths = 2^(4:6), testalpha = 0.0567, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "jsmurfLR", testlengths = 2^(4:6), testalpha = 0.0567, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter) expect_identical(critVal(n = 64L, alpha = 0.1, stat = teststat, family = "jsmurfLR", filter = testfilter), critVal(n = 64L, nq = 64L, alpha = 0.1, stat = teststat, family = "jsmurfLR", filter = testfilter)) testVector(teststat = teststat, testn = 64L, testnq = 128L, testfamily = "jsmurfLR", testlengths = 2^(4:6), testalpha = 0.01, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 67L, testfamily = "jsmurfLR", testlengths = 2^(4:6), testalpha = 0.01, testweights = rep(1 / 3, 3), defaultPenalty = "sqrt", filter = testfilter) expect_identical(critVal(n = 64L, alpha = 0.1, stat = teststat, family = "jsmurfLR", filter = testfilter), critVal(n = 64L, penalty = "sqrt", alpha = 0.1, stat = teststat, family = "jsmurfLR", filter = testfilter)) teststat <- monteCarloSimulation(n = 32L, r = 100L, family = "jsmurf", filter = testfilter) expect_identical(critVal(n = 32L, alpha = 0.1, stat = teststat, family = "jsmurf", filter = testfilter), critVal(n = 32L, alpha = 0.1, stat = teststat, family = "jsmurf", filter = testfilter, intervalSystem = "dyaLen")) teststat <- monteCarloSimulation(n = 16L, r = 100L, family = "jsmurfPS", filter = testfilter, intervalSystem = "all") testVector(teststat = teststat, testn = 16L, testnq = 20L, testlengths = c(13, 15:16), compareStat = teststat[c(2, 4:5), ], testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 3, 3), testintervalsystem = "all", testfamily = "jsmurfPS", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 16L, testnq = 20L, testlengths = 12:16, testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "all", testfamily = "jsmurfPS", filter = testfilter) teststat <- monteCarloSimulation(n = 64L, r = 100L, family = "jsmurf", filter = testfilter, intervalSystem = "dyaPar") testVector(teststat = teststat, testn = 64L, testnq = 64L, testlengths = 2^(4:6), testalpha = 0.1, defaultPenalty = "sqrt", testweights = rep(1 / 3, 3), testintervalsystem = "dyaPar", testfamily = "jsmurf", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 99L, testlengths = 2^(5:6), testalpha = 0.1, defaultPenalty = "sqrt", testintervalsystem = "dyaPar", testfamily = "jsmurf", filter = testfilter) compare <- 1 attr(compare, "n") <- 64L expect_identical(critVal(q = 1, n = 64L, output = "value", family = "jsmurfLR", filter = testfilter), compare) compare <- rep(1, 2) attr(compare, "n") <- 32L expect_identical(critVal(q = rep(1, 2), n = 32L, family = "jsmurfPS", filter = testfilter), compare) testq <- 1 attr(testq, "n") <- 30L compare <- 1 attr(compare, "n") <- 30L expect_identical(critVal(q = testq, n = 20L, stat = teststat, output = "value", family = "jsmurf", filter = testfilter), compare) testq <- as.numeric(1:64) attr(testq, "n") <- 64L compare <- as.numeric(c(16, 32)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, family = "jsmurfPS", filter = testfilter), compare) testq <- as.numeric(2^(4:6)) attr(testq, "n") <- 64L compare <- as.numeric(2^(4:5)) attr(compare, "n") <- 64L expect_identical(critVal(q = testq, n = 32L, family = "jsmurf", filter = testfilter), compare) teststat <- monteCarloSimulation(64L, r = 100L, output = "maximum", intervalSystem = "dyaPar", lengths = c(32L, 64L), penalty = "log", family = "jsmurfPS", filter = testfilter) expect_identical(critVal(64L, alpha = 0.1, r = 100L, intervalSystem = "dyaPar", lengths = c(32L, 64L), penalty = "log", family = "jsmurfPS", filter = testfilter, options = list(simulation = "vector", save = list(), load = list())), critVal(64L, alpha = 0.1, stat = teststat, intervalSystem = "dyaPar", lengths = c(32L, 64L), penalty = "log", family = "jsmurfPS", filter = testfilter)) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = c(32L, 64L), family = "jsmurfPS", filter = testfilter) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = c(32L, 64L), options = list(simulation = "vectorIncreased", save = list(), load = list()), family = "jsmurfPS", filter = testfilter), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = c(32L, 64L), family = "jsmurfPS", filter = testfilter)) teststat <- monteCarloSimulation(120L, r = 100L, output = "maximum", lengths = c(16, 32, 64), family = "jsmurf", filter = testfilter) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = c(16, 32, 64), nq = 120L, family = "jsmurf", filter = testfilter, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = c(16, 32, 64), family = "jsmurf", filter = testfilter)) teststat <- monteCarloSimulation(100L, r = 100L, family = "jsmurfPS", filter = testfilter) expect_identical(critVal(100L, alpha = 0.1, r = 100L, lengths = c(16, 64), family = "jsmurfPS", filter = testfilter, options = list(simulation = "matrix", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, lengths = c(16, 64), family = "jsmurfPS", filter = testfilter)) teststat <- monteCarloSimulation(127L, r = 100L, intervalSystem = "dyaLen", family = "jsmurfLR", filter = testfilter) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(32, 64), family = "jsmurfLR", filter = testfilter, options = list(simulation = "matrixIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(32, 64), family = "jsmurfLR", filter = testfilter)) testStepR <- new.env() teststat <- monteCarloSimulation(100L, r = 100L, family = "jsmurf", filter = testfilter) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "vector", family = "jsmurf", filter = testfilter, options = list(simulation = "matrix", save = list(workspace = "matrix"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", family = "jsmurf", filter = testfilter, options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "vector", family = "jsmurf", filter = testfilter, options = list(simulation = "matrix", save = list(), load = list(workspace = "matrix"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", options = list(), family = "jsmurf", filter = testfilter)) remove(critValStepRTab, envir = testStepR) }) test_that("family 'hjsmurf' works", { testthat::skip_on_cran() testfilter <- lowpassFilter::lowpassFilter(type = "bessel", param = list(pole = 4L, cutoff = 0.1), sr = 1e4) teststat <- monteCarloSimulation(n = 64L, r = 100L, family = "hjsmurf", filter = testfilter) expect_error(critVal(stat = teststat, family = "hjsmurf", filter = testfilter)) expect_warning(ret <- critVal(n = 64L, stat = teststat, family = "hjsmurf", filter = testfilter)) expect_identical(ret, critVal(alpha = 0.5, n = 64L, stat = teststat, nq = 64L, family = "hjsmurf", filter = testfilter, intervalSystem = "dyaLen", lengths = 2^(4:6), penalty = "weights", output = "vector")) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "hjsmurf", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "weights", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "hjsmurf", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "weights", filter = testfilter) teststat <- monteCarloSimulation(n = 64L, r = 100L, family = "hjsmurfSPS", filter = testfilter) expect_error(critVal(stat = teststat, family = "hjsmurfSPS", filter = testfilter)) expect_warning(ret <- critVal(n = 64L, stat = teststat, family = "hjsmurfSPS", filter = testfilter)) expect_identical(ret, critVal(alpha = 0.5, n = 64L, stat = teststat, nq = 64L, family = "hjsmurfSPS", filter = testfilter, intervalSystem = "dyaLen", lengths = 2^(4:6), penalty = "weights", output = "vector")) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "hjsmurfSPS", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "weights", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "hjsmurfSPS", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "weights", filter = testfilter) teststat <- monteCarloSimulation(n = 64L, r = 100L, family = "hjsmurfLR", filter = testfilter) expect_error(critVal(stat = teststat, family = "hjsmurfLR", filter = testfilter)) expect_warning(ret <- critVal(n = 64L, stat = teststat, family = "hjsmurfLR", filter = testfilter)) expect_identical(ret, critVal(alpha = 0.5, n = 64L, stat = teststat, nq = 64L, family = "hjsmurfLR", filter = testfilter, intervalSystem = "dyaLen", lengths = 2^(4:6), penalty = "weights", output = "vector")) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "hjsmurfLR", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "weights", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "hjsmurfLR", testlengths = 2^(4:6), testalpha = 0.1, testweights = rep(1 / 3, 3), defaultPenalty = "weights", filter = testfilter) testVector(teststat = teststat, testn = 64L, testnq = 124L, testfamily = "hjsmurfLR", testlengths = 2^(4:6), testalpha = 0.01, testweights = rep(1 / 3, 3), defaultPenalty = "weights", filter = testfilter, shift = -testfilter$len) testValue(teststat = teststat, testn = 64L, testnq = 124L, testfamily = "hjsmurfLR", testlengths = 2^(4:6), testalpha = 0.01, testweights = rep(1 / 3, 3), defaultPenalty = "weights", filter = testfilter) }) test_that("family 'LR' works", { testthat::skip_on_cran() testfilter <- lowpassFilter::lowpassFilter(type = "bessel", param = list(pole = 4L, cutoff = 0.1), sr = 1e4) teststat <- monteCarloSimulation(n = 64L, r = 100L, family = "LR", filter = testfilter) expect_error(critVal(stat = teststat, family = "LR", filter = testfilter)) expect_warning(ret <- critVal(n = 64L, stat = teststat, family = "LR", filter = testfilter)) expect_identical(ret, critVal(alpha = 0.5, n = 64L, stat = teststat, nq = 64L, family = "LR", filter = testfilter, intervalSystem = "all", lengths = 1:20, penalty = "weights", output = "vector")) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "LR", testlengths = 1:20, testalpha = 0.1, testweights = rep(1 / 20, 20), defaultPenalty = "weights", filter = testfilter) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "LR", testlengths = 1:20, testalpha = 0.1, testweights = rep(1 / 20, 20), defaultPenalty = "weights", filter = testfilter) expect_error(critVal(q = "s", alpha = 0.1, n = 64L, stat = teststat, family = "LR", filter = testfilter)) expect_error(critVal(q = rep(1, 6), alpha = 0.1, n = 64L, stat = teststat, family = "LR", filter = testfilter)) expect_error(critVal(alpha = 0, n = 64L, stat = teststat, family = "LR", filter = testfilter)) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "LR", testlengths = 1:20, testalpha = 0.0567, testweights = rep(1 / 20, 20), defaultPenalty = "weights", filter = testfilter) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "LR", testlengths = 1:20, testalpha = 0.0567, testweights = rep(1 / 20, 20), defaultPenalty = "weights", filter = testfilter) testVector(teststat = teststat, testn = 64L, testnq = 123L, testfamily = "LR", compareStat = teststat[1:10, ], testlengths = 1:10, testalpha = 0.015, testweights = 1:10 / sum(1:10), defaultPenalty = "weights", filter = testfilter) testValue(teststat = teststat, testn = 64L, testnq = 128L, testfamily = "LR", testlengths = c(1L, 5L, 18L), testalpha = 0.05, testweights = c(3, 5, 8) / sum(c(3, 5, 8)), defaultPenalty = "weights", filter = testfilter) teststat <- monteCarloSimulation(n = 128L, r = 100L, family = "LR", filter = testfilter, intervalSystem = "dyaLen") testVector(teststat = teststat, testn = 128L, testnq = 200L, testlengths = c(2L, 4L, 16L), compareStat = teststat[c(2, 3, 5), ], testalpha = 0.1, defaultPenalty = "weights", testweights = rep(1 / 3, 3), testintervalsystem = "dyaLen", testfamily = "LR", filter = testfilter) testValue(teststat = teststat, testn = 128L, testnq = 200L, testlengths = c(2L, 4L, 16L), testalpha = 0.1, defaultPenalty = "weights", testintervalsystem = "dyaLen", testfamily = "LR", filter = testfilter) compare <- rep(1, 16) attr(compare, "n") <- 16L expect_identical(critVal(q = rep(1, 16), n = 16L, stat = teststat, family = "LR", filter = testfilter), compare) testq <- rep(1, 16) attr(testq, "n") <- 20L compare <- rep(1, 16) attr(compare, "n") <- 20L expect_identical(critVal(q = testq, n = 16L, stat = teststat, family = "LR", filter = testfilter), compare) testq <- as.numeric(2^(0:4)) attr(testq, "n") <- 16L compare <- as.numeric(2^(0:4)) attr(compare, "n") <- 16L expect_identical(critVal(q = testq, n = 16L, family = "LR", filter = testfilter, intervalSystem = "dyaLen"), compare) expect_error(critVal(100L, alpha = 0.1, r = 100L, intervalSystem = "dyaPar", family = "LR", filter = testfilter)) teststat <- monteCarloSimulation(127L, r = 100L, output = "maximum", lengths = 2:30) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", lengths = 2:30, options = list(simulation = "vectorIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", lengths = 2:30)) teststat <- monteCarloSimulation(100L, r = 100L, family = "LR", filter = testfilter) expect_identical(critVal(100L, alpha = 0.1, r = 100L, lengths = c(2, 8, 16), family = "LR", filter = testfilter, options = list(simulation = "matrix", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, lengths = c(2, 8, 16), family = "LR", filter = testfilter)) teststat <- monteCarloSimulation(127L, r = 100L, intervalSystem = "dyaLen", family = "LR", filter = testfilter) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 16), family = "LR", filter = testfilter, options = list(simulation = "matrixIncreased", save = list(), load = list())), critVal(100L, alpha = 0.1, stat = teststat, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 16), family = "LR", filter = testfilter)) expect_error(critVal(100L, alpha = 0.1, stat = teststat, output = "value", intervalSystem = "dyaLen", penalty = "log", lengths = c(4, 8, 32), family = "LR", filter = testfilter)) testStepR <- new.env() teststat <- monteCarloSimulation(100L, r = 100L, family = "LR", filter = testfilter) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "vector", family = "LR", filter = testfilter, options = list(simulation = "matrix", save = list(workspace = "matrix"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", family = "LR", filter = testfilter, options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(100L, alpha = 0.1, r = 50L, output = "vector", family = "LR", filter = testfilter, options = list(simulation = "matrix", save = list(), load = list(workspace = "matrix"), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", options = list(), family = "LR", filter = testfilter)) remove(critValStepRTab, envir = testStepR) testStepR <- new.env() teststat <- monteCarloSimulation(100L, r = 100L, family = "LR", localVal = mean, filter = testfilter) expect_identical(critVal(100L, alpha = 0.1, r = 100L, output = "vector", family = "LR", filter = testfilter, localValue = mean, options = list(simulation = "matrix", save = list(workspace = "matrix"), load = list(), envir = testStepR)), critVal(100L, alpha = 0.1, stat = teststat, output = "vector", family = "LR", filter = testfilter, options = list())) expect_false(exists("critValStepRTab", envir = testStepR, inherits = FALSE)) }) test_that("family '2Param' works", { testthat::skip_on_cran() testfilter <- lowpassFilter::lowpassFilter(type = "bessel", param = list(pole = 4L, cutoff = 0.1), sr = 1e4) teststat <- monteCarloSimulation(n = 64L, r = 2L, family = "2Param", filter = testfilter) expect_error(critVal(stat = teststat, family = "2Param", filter = testfilter)) expect_warning(ret <- critVal(n = 64L, stat = teststat, family = "2Param", filter = testfilter)) expect_identical(ret, critVal(alpha = 0.5, n = 64L, stat = teststat, nq = 64L, family = "2Param", filter = testfilter, intervalSystem = "all", lengths = 1:64, penalty = "weights", output = "vector")) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "2Param", testlengths = 1:64, testalpha = 0.1, testweights = rep(1 / 64, 64), defaultPenalty = "weights", filter = testfilter) testVector(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "2Param", compareStat = teststat[c(1, 3:5, 16),], testlengths = c(1, 3:5, 16), testalpha = 0.1, testweights = rep(1 / 5, 5), defaultPenalty = "weights", filter = testfilter) testValue(teststat = teststat, testn = 64L, testnq = 64L, testfamily = "2Param", testlengths = c(1, 3:5, 16), testalpha = 0.1, testweights = rep(1 / 5, 5), defaultPenalty = "weights", filter = testfilter) testVector(teststat = teststat, testn = 64L, testnq = 164L, testfamily = "2Param", compareStat = teststat[c(1, 3:5, 16),], testlengths = c(1, 3:5, 16), testalpha = 0.1, testweights = rep(1 / 5, 5), defaultPenalty = "weights", filter = testfilter) testValue(teststat = teststat, testn = 64L, testnq = 127L, testfamily = "2Param", testlengths = c(1, 3:5, 16), testalpha = 0.1, testweights = rep(1 / 5, 5), defaultPenalty = "weights", filter = testfilter) testStepR <- new.env() expect_identical(critVal(64L, alpha = 0.1, r = 2L, output = "vector", family = "2Param", filter = testfilter, options = list(simulation = "matrix", save = list(workspace = "matrix"), load = list(), envir = testStepR)), critVal(64L, alpha = 0.1, stat = teststat, output = "vector", family = "2Param", filter = testfilter, options = list())) expect_identical(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat[[1]], teststat) expect_identical(length(get("critValStepRTab", envir = testStepR, inherits = FALSE)$stat), 1L) expect_identical(critVal(64L, alpha = 0.1, r = 1L, output = "vector", family = "2Param", filter = testfilter, lengths = c(1, 3:5, 16), messages = 1L, options = list(simulation = "matrix", save = list(), load = list(workspace = "matrix"), envir = testStepR)), critVal(64L, alpha = 0.1, stat = teststat, output = "vector", options = list(), family = "2Param", filter = testfilter, lengths = c(1, 3:5, 16))) remove(critValStepRTab, envir = testStepR) testStepR <- new.env() ret <- critVal(64L, alpha = 0.1, r = 2L, output = "vector", family = "2Param", filter = testfilter, localVar = function(data) 1, options = list(simulation = "matrix", save = list(workspace = "matrix"), load = list(), envir = testStepR), lengths = c(1, 3:5, 16)) expect_false(exists("critValStepRTab", envir = testStepR, inherits = FALSE)) }) test_that("everything is working if data$lengths is not in data$defaultLengths", { testdata <- rnorm(100) testfilter <- lowpassFilter::lowpassFilter(type = "bessel", param = list(pole = 4L, cutoff = 0.1), sr = 1, len = 8L, shift = 0.5) ret <- critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = 21:23) expect_identical(length(ret), 3L) expect_identical(attr(ret, "n"), 127L) ret <- critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21)) expect_identical(length(ret), 11L) expect_identical(attr(ret, "n"), 127L) ret <- critVal(n = length(testdata), family = "LR", filter = testfilter, q = 1, lengths = c(1:10, 21), penalty = "sqrt") expect_identical(length(ret), 11L) expect_identical(attr(ret, "n"), 100L) for (i in 1:length(ret)) { ret[i] <- penalisationSqrt(ret[i], 100, c(1:10, 21)[i]) } expect_equal(ret, rep(ret[1], length(ret)), check.attributes = FALSE) teststat <- monteCarloSimulation(n = 100L, r = 10L, family = "LR", filter = testfilter) expect_error(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, stat = teststat, lengths = c(1:10, 21))) teststat <- monteCarloSimulation(n = 127L, r = 10L, family = "LR", filter = testfilter, lengths = c(1:10, 21), output = "maximum", penalty = "sqrt") expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt"), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", output = "value"), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat, output = "value")) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "vectorIncreased")), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) teststat <- monteCarloSimulation(n = 100L, r = 10L, family = "LR", filter = testfilter, lengths = c(1:10, 21), output = "maximum", penalty = "sqrt") expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "matrix")), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "matrix"), output = "value"), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat, output = "value")) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "vector")), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "matrixIncreased"), nq = 100L), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) ret <- critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat, output = "value") expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, q = ret, penalty = "sqrt", lengths = c(1:10, 21)), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) testfile <- c(tempfile(pattern = "file", tmpdir = tempdir(), fileext = ".RDS"), tempfile(pattern = "file", tmpdir = tempdir(), fileext = ".RDS")) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "matrix", save = list(RDSfile = testfile))), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) attr(teststat, "save") <- FALSE expect_identical(readRDS(testfile[1]), teststat) expect_identical(class(readRDS(testfile[2]))[1], "MCSimulationVector") expect_identical(attr(readRDS(testfile[2]), "lengths"), c(1:10, 21L)) expect_identical(dim(readRDS(testfile[2])), c(11L, 10L)) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "matrix", load = list(RDSfile = testfile[1]))), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "matrix", load = list(RDSfile = testfile[2]))), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) unlink(testfile) testvariable <- c("testsavevector", "testsavematrix") testStepR <- new.env() expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "matrix", save = list(variable = testvariable), envir = testStepR)), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) expect_identical(get("testsavevector", envir = testStepR), teststat) expect_identical(class(get("testsavematrix", envir = testStepR))[1], "MCSimulationVector") expect_identical(attr(get("testsavematrix", envir = testStepR), "lengths"), c(1:10, 21L)) expect_identical(dim(get("testsavematrix", envir = testStepR)), c(11L, 10L)) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = get("testsavematrix", envir = testStepR)), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) remove("testsavevector", envir = testStepR) remove("testsavematrix", envir = testStepR) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "vector", save = list(variable = testvariable), envir = testStepR)), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) attr(teststat, "save") <- TRUE expect_identical(get("testsavevector", envir = testStepR), teststat) expect_false(exists("testsavematrix", envir = testStepR)) remove(testsavevector, envir = testStepR) testStepR <- new.env() expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "matrix", save = list(workspace = c("matrix", "vector", "matrixIncreased", "vectorIncreased")), envir = testStepR)), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) expect_false(exists("critValStepRTab", where = testStepR, inherits = FALSE)) expect_identical(critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(1:10, 21), penalty = "sqrt", options = list(simulation = "matrix", save = list(fileSystem = c("matrix", "vector", "matrixIncreased", "vectorIncreased")), dirs = "testStepR")), critVal(n = length(testdata), family = "LR", filter = testfilter, alpha = 0.05, lengths = c(1:10, 21), penalty = "sqrt", stat = teststat)) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) ret <- critVal(n = length(testdata), family = "2Param", filter = testfilter, alpha = 0.05, r = 2L, lengths = 70:71) expect_identical(length(ret), 2L) expect_identical(attr(ret, "n"), 127L) teststat <- monteCarloSimulation(n = 100L, r = 2L, family = "2Param", filter = testfilter) expect_error(critVal(n = length(testdata), family = "2Param", filter = testfilter, alpha = 0.05, stat = teststat, lengths = c(10, 70))) teststat <- monteCarloSimulation(n = 127L, r = 10L, family = "2Param", filter = testfilter, lengths = c(10, 75), output = "maximum", penalty = "sqrt") expect_identical(critVal(n = length(testdata), family = "2Param", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(10, 75), penalty = "sqrt"), critVal(n = length(testdata), family = "2Param", filter = testfilter, alpha = 0.05, lengths = c(10, 75), penalty = "sqrt", stat = teststat)) expect_identical(critVal(n = length(testdata), family = "2Param", filter = testfilter, alpha = 0.05, r = 10L, lengths = c(10, 75), penalty = "sqrt", nq = 127L, options = list(simulation = "matrixIncreased", save = list(fileSystem = c("matrix", "vector", "matrixIncreased", "vectorIncreased")), dirs = "testStepR")), critVal(n = length(testdata), family = "2Param", filter = testfilter, alpha = 0.05, lengths = c(10, 75), penalty = "sqrt", stat = teststat)) expect_false(file_test(op = "-d", file.path(R.cache::getCacheRootPath(), "testStepR"))) }) test_that("results of digest are as expected", { expect_identical(digest::sha1("gauss"), "55f46ed9e4d5824c6dbbb5dd42d3e5612aa38c38") expect_identical(digest::sha1("hsmuce"), "ea85b0b1784732506a5dadd685474b19542d5499") expect_identical(digest::sha1(list("mDependentPS", c(0.8, 0.5, 0.3)), digits = 6), "bafaebb5df35ee66e46fbd34a6f8c34027600481") expect_identical(digest::sha1(list("jsmurf", c(0.8, 0.5, 0.3)), digits = 6), "332970dee30c4832bc1661232188aede0ed3baef") expect_identical(digest::sha1(list("jsmurfPS", c(0.8, 0.5, 0.3)), digits = 6), "8ec71194bd3246a4ee6213a003b4799afda0e74c") expect_identical(digest::sha1(list("jsmurfLR", c(0.8, 0.5, 0.3)), digits = 6), "648798c249f8e2530db631c0f56e46b310a20bb0") expect_identical(digest::sha1(list("hjsmurf", c(0.8, 0.5, 0.3)), digits = 6), "607c95ac759e448e399047206a175ff2010a4f21") expect_identical(digest::sha1(list("hjsmurfSPS", c(0.8, 0.5, 0.3)), digits = 6), "89d2d4235b32a2f3b57bb046aec805a886427d71") expect_identical(digest::sha1(list("hjsmurfLR", c(0.8, 0.5, 0.3)), digits = 6), "4df7ccdbdf72cbe02223f1587f882b78e1e12b40") expect_identical(digest::sha1(list("2Param", "bessel", unname(list(pole = 4L, cutoff = 0.1)), 1e4, 11L), digits = 6), "b08645fa99d317fb3c695af2d287f20dc4c61199") expect_identical(digest::sha1(list("LR", "bessel", unname(list(pole = 4L, cutoff = 0.1)), 1e4, 11L), digits = 6), "224d015261c8749de186db44add9452c7ea8ba39") expect_identical(digest::digest(list(31, digest::sha1("gauss"), "all")), "19d55f7c6d6b9db86f1b4f1123fbc9b6") expect_identical(digest::digest(list(120, digest::sha1("hsmuce"), "dyaPar")), "8bb33d5ae7ce51a591342a8ebfa9b83f") expect_identical(digest::digest(list(19, digest::sha1(list("mDependentPS", c(0.8, 0.5, 0.3))), "dyaLen")), "b29ba0b06d7e287384cbb20ed22a1ac6") expect_identical(digest::digest(list(6e5, digest::sha1(list("LR", "bessel", unname(list(pole = 4L, cutoff = 0.1)), 1e4, 11L), digits = 6), "dyaLen")), "be51aa263b613b97498b285c59f6ea0a") expect_identical(digest::digest(list(31, digest::sha1("gauss"), "all", 1:31, "sqrt")), "3ae1d6f7ad6615143286979e6f43dab2") expect_identical(digest::digest(list(120, digest::sha1("hsmuce"), "dyaPar", c(2, 4, 16), "none")), "d872fa9a922a1898076cb3939e04ef96") expect_identical(digest::digest(list(19, digest::sha1(list("mDependentPS", c(0.8, 0.5, 0.3))), "dyaLen", c(2, 4, 8, 16), "log")), "7f02f681cef8e209c54e2b4f303e8a71") expect_identical(digest::digest(list(3e6, digest::sha1(list("2Param", "bessel", unname(list(pole = 4L, cutoff = 0.1)), 1e4, 11L), digits = 6), "dyaLen", c(2, 4, 8, 16), "weights")), "ff33d9102cd0cfc1091ad70079342c49") })