Add tests using testthat

[nikolase90]

If you're working on one of these files, please add the url to your branch or the pull request. Mark the box if the changes are merged with master.

R-files

• clustering.R
• explanation.RTests for shapley.R and explanation.R #128
• features.R Splits feature_combinations into multiple functions #70
• observations.R Tests for R/observations.R #95
• plot.R Tests for plot function #86
• predictions.R Unit tests for R/predictions.R #109
• sampling.RTests for functions in R/sampling.R (#91) #92
• shapley.R Tests for shapley.R and explanation.R #128
• transformation.R Tests for R/transformation.R #77
• utils.R Currently there are zero functions in R/utils.R.
• models.R Unit tests and fixes for R/models.R #129

src-files

• AICc.cpp
• distance.cpp
• impute_data.cpp Adds tests for function in src/impute_data.cpp #78
• weighted_matrix.cpp Adds tests for functions in src/weighted_matrix.cpp #73

The following files should not be tested

• R/shapr-package.R
• R/zzz.R

[martinju]

I am getting a bit worried that every time we make a change to shap or explain functions, we have to re-create the test objects (explanation_explain_obj.rds and similar) which compares the results of the previous run to pass the tests. That means we don't have a good test for this. This is especially dangerous if one does many changes in the same PR (although we seldom do that).

My suggestion is to add a seperate test that checks only the shapley values from the different test models. We could run a lapply call to the ex_list, extract only the data.table with the shapley values, and then do a testthat::expect_known_value or testthat::expect_known_object on that. What do you think @nikolase90 ? Or is there a better option?

[nikolase90]

@martinju I think that sounds like a good way to do it.