context("Test: REMatrix() ") data(PhyloExpressionSetExample) data(DivergenceExpressionSetExample) equal_df <- function(df1, df2) { rownames(df1) <- NULL rownames(df2) <- NULL isTRUE(all.equal(df1, df2)) } TestREMatrix1 <- t(apply(dplyr::summarise_all( dplyr::group_by(PhyloExpressionSetExample[, c(1, 3:9)], Phylostratum), list(mean) )[, 2:8], 1, function(x) (x - min(x)) / (max(x) - min(x)))) rownames(TestREMatrix1) <- 1:12 TestREMatrix2 <- t(apply(dplyr::summarise_all( dplyr::group_by(DivergenceExpressionSetExample[, c(1, 3:9)], Divergence.stratum), list(mean) )[, 2:8], 1, function(x) (x - min(x)) / (max(x) - min(x)))) rownames(TestREMatrix2) <- 1:10 nonStandardExpressionSet <- PhyloExpressionSetExample[, 2:9] test_that("is.ExpressionSet() throughs error when no ExpressionSet is entered to REMatrix()", { expect_error( REMatrix(nonStandardExpressionSet) ) }) test_that("REMatrix() computes correct values...", { expect_true(equal_df(REMatrix(PhyloExpressionSetExample), TestREMatrix1)) expect_true(equal_df( REMatrix(DivergenceExpressionSetExample), TestREMatrix2 )) }) test_that("error occurs when when REMatrix() is computed on only one developmental stage ", { expect_error(REMatrix(PhyloExpressionSetExample[, 1:3])) })