devel stuff: notes for test update (bladen)

"C:/Users/Work/Desktop/mO Work/Test Ground Truths//.RData"

Test types:

• basic
  • just tests the simplest use case works for ALL possible input objects
• data
  • similar to basic, but uses different input datasets
• parameter
  • tests the functionality of a specific parameter (or set of parameters)
• edge case
  • tests for warnings or odd scenarios
• error
  • tests that a specific error is raised in the appropriate scenario

Things to test for:

• names of output ocmponents
• dimensions of output
• test full dataframes numerically (explore this a bit)
• test pass through of input parameters
• test type of output components
• test when certain errors should be raised

What do we already have:

• auroc
  • only test for comp auroc values on one dataset
  • different parameters, different datasets, test for more values
• background_predict
  • only test one dataset and only test two numerical values.
  • different parameters, different datasets, test for more values
• cim
  • tests for matricies, rcc, spca, spls (x2) and multilevel. two datasets. fairly sufficient for now
  • could maybe test with a variety of different parameters
• circosPlot
  • two datasets, only test that circos "works" (tests output is of matrix type)
  • check different parameter usage and wider range of numerical testing
• diablo
  • wide range of tests but only one dataset. small number of numerical values evaluated
  • different parameters, another dataset
• internals-.get-pch
  • doesn't use any mixOmics datasets
  • explore different parameters and error states
• internals
  • tests .get.ind.colors,.are.colors, .get.colors, .get.character.vector, .check_test.keepX and .check_ncomp
  • see if there are more internals to test, explore different datasets with these tests
• network
  • two datasets, only 1 numerical value for each
  • variety of parameters to be tested
• pca
  • only one test with one dataset and no numerical values are assessed, a lot of work here
• perf.diablo
  • only one dataset, need to test edge cases (low numbers of repeats/folds - catch errors)\
  • test wider range of the numerical output
• perf.mint.splsda
  • only tests choice.ncomp and how alpha affects things. much work here
• plotIndiv
  • one of the better ones - though only one or two values tested for each case, maybe just a few few parameter tests and any remaining methods (ie pca) and expand on how many components each test checks
  • test for rcc x2, (s)pls x2, (s)plsda, mint.(s)plsda, sgcca x2 and sgccada x3
  • 4 datasets
• plotLoadings
  • tests spls, splsda, block.splsda, mint.splsda. 3 datasets
  • add some more parameter tests + more numerical values
• plotVar
  • needs tons of work, only one test
• predict
  • 4 datasets, mint.splsda, block.splsda, pls, plsda
  • add any remaining methods (eg splsda) and test for more parameters
• tune.block.splsda
  • only one test (one dataset), no parameter tests
• tune.mint.splsda
  • two tests, one dataset. test more parameters (not just signif.threshold)
• tune.spls
  • one test, one dataset, with/without parallel
  • needs work
• tune.splsda
  • one test, one dataset

What do we need new test files for:

Front end functions:

• biplot
• block.(s)pls
• block.(s)plsda
• cimDiablo
• ipca
• mint.block.(s)pls
• mint.block.(s)plsda
• mint.pca
• mint.(s)pls
• mint.(s)plsda
• network
• nipals
• all perf variants
• all plot variants
• all plot.tune variants
• plotMarkers
• (s)pls
• (s)plsda
• rcc
• selectVar
• sipca
• spca
• all tune variants (pca, rcc, spca, splslevel)
• wrapper.rgcca
• wrapper.sgcca

Back end functions:

• explained_variance
• get.confusion_matrix
• impute.nipals
• logratio-transformations
• map
• nearZeroVar
• study_split
• unmap
• vip
• withinVariation