# Copyright (C) 2013 - 2022 Metrum Research Group # # This file is part of mrgsolve. # # mrgsolve is free software: you can redistribute it and/or modify it # under the terms of the GNU General Public License as published by # the Free Software Foundation, either version 2 of the License, or # (at your option) any later version. # # mrgsolve is distributed in the hope that it will be useful, but # WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with mrgsolve. If not, see . library(testthat) library(mrgsolve) library(dplyr) Sys.setenv(R_TESTS="") options("mrgsolve_mread_quiet"=TRUE) mod <- mrgsolve::house() %>% update(end=240) context("test-data_set") data(extran1) lo <- extran1 up <- extran1 names(up) <- toupper(names(up)) test_that("Same result from upper and lower case names", { what <- c(1,5:7) a <- mod %>% data_set(up) %>% mrgsim b <- mod %>% data_set(lo) %>% mrgsim expect_identical(as.matrix(a)[,what],as.matrix(b)[,what]) expect_identical(a$TIME,b$time) expect_is(plot(a),"trellis") expect_is(plot(b),"trellis") expect_is(plot(a, CP~.),"trellis") expect_is(plot(a, CP~TIME),"trellis") expect_is(plot(b, CP~.),"trellis") expect_is(plot(b, CP~time),"trellis") expect_error(plot(b, CP~TIME)) ldd <- expand.ev(amt=100, ii=24, addl=2, ss=1) udd <- ldd names(udd) <- toupper(names(udd)) a <- mod %>% data_set(ldd) %>% mrgsim b <- mod %>% data_set(udd) %>% mrgsim expect_identical(as.matrix(a)[,what],as.matrix(b)[,what]) }) test_that("Warning is generated when mixed upper/lower names", { mix <- dplyr::rename(lo, EVID = evid) expect_warning(mrgsim(data_set(mod,mix))) }) test_that("Include TIME and time when checking for mixed upper/lower case", { expect_warning( mrgsim(house(), data = data.frame(ID = 1, TIME = 0, amt = 0, cmt = 1)), "Both lower- & upper-case", fixed = TRUE ) expect_warning( mrgsim(house(), data = data.frame(ID = 1, time = 0, AMT = 0, CMT = 1)), "Both lower- & upper-case", fixed = TRUE ) }) test_that("Filter out ID", { out <- mod %>% data_set(up, ID > 4) %>% mrgsim expect_true(all(out$ID > 4)) }) test_that("ID is required", { df <- expand.ev(amt=100,ii=12,addl=2) %>% dplyr::select(-ID) expect_error(mod %>% data_set(df) %>% mrgsim) }) test_that("cmt is required", { df <- expand.ev(amt=100,ii=12,addl=2) %>% dplyr::select(-cmt) expect_error(mod %>% data_set(df) %>% mrgsim) }) test_that("time is required", { df <- expand.ev(amt=100,ii=12,addl=2) %>% dplyr::select(-time) expect_error(mod %>% data_set(df) %>% mrgsim) }) mod <- mrgsolve::house() data(exTheoph) dd1 <- exTheoph %>% bind_rows(., group_by(.,ID) %>% slice(5)) %>% arrange(ID) dd2 <- dd1 %>% arrange(ID,time) test_that("Improperly sorted records produces error", { expect_error(mod %>% data_set(dd1) %>% mrgsim) }) test_that("Properly sorted records produces no error", { expect_is((mod %>% data_set(dd2) %>% mrgsim),"mrgsims") }) test_that("Data set column order gives same answer", { mod <- mrgsolve::house() %>% omat(dmat(1,1,1,1)) data(extran3) set.seed(9923403) extran3 <- extran3 %>% mutate(a=-10, Z = -11, M = -22, h = -33) extran3b <- extran3[,sample(seq_along(names(extran3)))] set.seed(22930) out1 <- mod %>% carry_out(a,m,M,h) %>% data_set(extran3) %>% mrgsim set.seed(22930) out2 <- mod %>% carry_out(a,m,M,h) %>% data_set(extran3b) %>% mrgsim expect_identical(as.matrix(out1),as.matrix(out2)) expect_false(all(names(extran3)==names(extran3b))) rm(mod) }) test_that("numerics_only", { n <- 10 data <- tibble( ID = as.numeric(seq(n)), EYES = "black", DATETM = as.POSIXct("2018-01-01 08:00:00"), DATE = as.Date(DATETM), INT = as.integer(ID), BOOL = INT > 6 ) df <- numerics_only(data, convert_lgl=TRUE) expect_equal(names(df), c("ID", "INT", "BOOL")) expect_true(all(mrgsolve:::numeric_columns(df))) expect_message(numerics_only(data)) df <- numerics_only(data,convert_lgl = FALSE) expect_equal(names(df), c("ID", "INT")) expect_true(all(mrgsolve:::numeric_columns(df))) expect_silent(numerics_only(data,quiet=TRUE)) }) test_that("missing value in param column is message", { data <- expand.ev(amt = 100, ii = 24, addl = 2, WT = NA_real_, FOO = 2) expect_warning(valid_data_set(data,mod), regexp="Parameter column WT must not contain missing values.") idata <- dplyr::select(data, ID,WT) expect_warning(valid_idata_set(idata,mod), regexp="Parameter column WT must not contain missing values.") }) test_that("missing value in time/rate/ID is error", { data <- data.frame(ID = 1, time = NA_real_, amt = 100, cmt = 1, evid=1) expect_error(mrgsim_d(mod,data), regexp="Found missing values in input data.") data <- data.frame(ID = NA_real_, rate = 100, amt = 100, cmt = 1, evid=1,time=0) expect_error(mrgsim_d(mod,data), regexp="Found missing values in input data.") }) test_that("observations expand", { e <- ev(amt = 100) dat <- expand_observations(e, c(1,2,3)) dose <- filter(dat,evid==1) expect_equal(nrow(dose),1) obs <- filter(dat, evid==0) expect_equal(nrow(obs),3) expect_equal(obs[["time"]],c(1,2,3)) }) test_that("expand observations bug issue-563", { e1 <- ev(amt = 100, cmt=1,LAGT=1,ID=15) e2 <- ev(time=4,amt=50,cmt=1,LAGT=0,rate=24,addl=2,ii=24,ID=15) e <- c(e1,e2) expect_silent(d <- expand_observations(e,seq(1,10))) expect_is(d,"data.frame") }) test_that("add position argument to expand observations issue-565", { e1 <- ev(amt = 100, ii = 4, addl = 1) %>% realize_addl dat1 <- expand_observations(e1,c(0,4)) dat2 <- expand_observations(e1,c(0,4),obs_pos=13) expect_equal(dat1$amt, c(0,100,0,100)) expect_equal(dat2$amt, c(100,0,100,0)) expect_equal(dat1$time,c(0,0,4,4)) expect_equal(dat1$time,dat2$time) }) test_that("Convert names to lower case with lctran [SLV-TEST-0004]", { data <- data.frame(time = 1, EVID = 2, ss = 2, foo = 5, BAR = 2) ans <- lctran(data) expect_equal( names(ans), c("time", "evid", "ss", "foo", "BAR") ) data <- data.frame(time = 1, EVID = 2, ss = 2, TIME = 5, BAR = 2) expect_warning( lctran(data), regexp = "There are both upper and lower case" ) }) test_that("Convert names to upper case with uctran [SLV-TEST-0005]", { data <- data.frame(time = 1, EVID = 2, ss = 2, foo = 5, BAR = 2) ans <- uctran(data) expect_equal( names(ans), c("TIME", "EVID", "SS", "foo", "BAR") ) data <- data.frame(time = 1, EVID = 2, ss = 2, TIME = 5, BAR = 2) expect_warning( uctran(data), regexp = "There are both upper and lower case" ) }) test_that("Convert event to upper or lower case [SLV-TEST-0006]", { x <- uctran(ev(amt = 100, ii = 5)) expect_equal(x@case, 1L) expect_is(x, "ev") x <- lctran(ev(amt = 100, ii = 5)) expect_equal(x@case, 0L) expect_is(x, "ev") })