Check and update data before explanation #245
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Currently not possible to test the custom model as get_supported_models searches globalenv, but apparently not the testing environment
and doing this checking within get_model_specs, but I don't require this function at all for custom models to simplify
aredelmeier
reviewed
Jan 22, 2021
aredelmeier
reviewed
Jan 22, 2021
aredelmeier
reviewed
Jan 22, 2021
aredelmeier
approved these changes
Jan 25, 2021
95% Annabelles comments. Some minor modifications Co-authored-by: Annabelle Redelmeier <[email protected]>
forgotten suggestion Co-authored-by: Annabelle Redelmeier <[email protected]>
martinju
commented
Jan 26, 2021
R/preprocess_data.R
Outdated
| # Reorder and delete unused columns | ||
| cnms_remove <- setdiff(colnames(data), new_labels) | ||
| if (length(cnms_remove) > 0) { | ||
| message(paste0("The columns(s) ",paste0(cnms_remove,collapse=", ")," is not used by the model and thus removed ", |
There was a problem hiding this comment.
It is just about avoiding too long lines. I am splitting the whole thing into a multiline ting now.
tests/testthat/test-explanation.R
Outdated
| prediction_zero = p0, sample = FALSE) | ||
| # Ex 18: Explain combined II - all empirical | ||
| approach <- c(rep("empirical", 4)) | ||
| ex_list[[18]] <- explain(x_test, explainer, approach = approach, prediction_zero = p0) |
There was a problem hiding this comment.
Good catch. Just a mistake. Will fix
tests/testthat/test-explanation.R
Outdated
| explain(x_test, explainer, approach = "ctree", prediction_zero = p0, sample = FALSE, | ||
| mc_cores_create_ctree = multicore, mc_cores_sample_ctree = 1) | ||
| ) | ||
| # Ex 38: Test that ctree with mincriterion equal to same probability four times gives the same as only passing one |
and do not accept NULL either
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Updated!
The point of this PR is to add proper checks that the data being passed to shapr (training data) and explain (testing data) have the correct format, meaning that they contain the right features, have the right class and if they are factor features have the right levels. While this was not crucial when we only supported numerical features, it has become important now that we support factors as well through the ctree approach. If there are differences between what is defined in model object and in the data, the data are either adjusted/updated and a message is provided, or an error is thrown.
All of this is mainly done through a the function
preprocess_data: The arguments to that function is the data to check and a list describing the "correct" feature details which is extracted from the model object using a new functionget_model_specswhich is an updated version of the previousfeaturesfunction (now obsolete, so deleted).preprocess_datacallsget_data_specswhich is a new function that extracts the feature info from the data. The new functioncheck_featuresis the main contribution, and that compares the feature list from the model with that from the data. That function is motivated by the earliermake_dummiesandapply_dummiesfunctions. The function does no re-shuffling, deletion of unused columns or so, but returns a feature_list that is passed on toupdate_datawhich handles that. The feature list is written to the explainer at the end ofshaprand passed on toexplainwhere the checking and updating is done for the test data.A bunch of remarks
get_model_specscalls the function get_supported_models which gives a table with all function classes defined for the functionsmodel_type,predict_model(andget_model_specs) which are needed to run explain a model class. This seems like a valid approach, but I currently have some issues testing custom models as testthat does not see functions that are defined within a testthat environment, and I am probably not allowed to write to .globalenv within the tests. Testing ofget_supported_modelsworks and is included in the standard test environment, while tests for custom models are for now done manually by runningexpect_silent(source("tests/testthat/manual_test_scripts/test_custom_models.R"))in the console. This works, but at some point we should find a better way to do this.The
make_dummiesfunction now also makes use of thecheck_featuresfunction to reuse more code.A full round of linting and styling of the full package will be done in a separate PR after this one, so just ignore all of that for now.
There are tests are not great, but covr::covr() shows that the majority of the code is tested, so I think it is fine for now. Ref
Restructure tests #249 , we should restucture the tests later, but not for this CRAN release.
Tests still failing on macOS. Don't understand why. Keep on ignoring for now.