Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Included warning TreeSHAP background dataset size. #710

Merged
merged 3 commits into from
Jul 4, 2022
Merged

Included warning TreeSHAP background dataset size. #710

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9b87e69
Included warning TreeSHAP background dataset size.
RobertSamoilescu Jun 29, 2022
db1c0e6
Fixed background size when DenseData object returned by summarisation.
RobertSamoilescu Jun 29, 2022
58ee40d
Updated waring to emphasize sampling with replacement.
RobertSamoilescu Jul 1, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 19 additions & 0 deletions alibi/explainers/shap_wrappers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1000,6 +1000,7 @@ def reset_predictor(self, predictor: Callable) -> None:
# TODO: Look into pyspark support requirements if requested
# TODO: catboost.Pool not supported for fit stage (due to summarisation) but can do if there is a user need

TREE_SHAP_BACKGROUND_SUPPORTED_SIZE = 100
TREE_SHAP_BACKGROUND_WARNING_THRESHOLD = 1000
TREE_SHAP_MODEL_OUTPUT = ['raw', 'probability', 'probability_doubled', 'log_loss']

Expand Down Expand Up @@ -1159,6 +1160,24 @@ def fit(self, # type: ignore[override]
else:
self._check_inputs(background_data)

# summarisation can return a DenseData object
n_samples = (background_data.data if isinstance(background_data, shap_utils.DenseData)
else background_data).shape[0]

# Warns the user that TreeShap supports only up to TREE_SHAP_BACKGROUND_SIZE(100) samples in the
# background dataset. Note that there is a logic above related to the summarisation of the background
# dataset which uses TREE_SHAP_BACKGROUND_WARNING_THRESHOLD(1000) as (warning) threshold. Although the
# TREE_SHAP_BACKGROUND_WARNING_THRESHOLD > TREE_SHAP_BACKGROUND_SUPPORTED_SIZE which is contradictory, we
# leave the logic above untouched. This approach has at least two benefits:
# i) minimal refactoring
# ii) return the correct result if a newer version of shap which fixes the issue is used before we
# update our dependencies in alibi (i.e. just ignore the warning)
if n_samples > TREE_SHAP_BACKGROUND_SUPPORTED_SIZE:
logger.warning(f'The upstream implementation of interventional TreeShap supports only up to '
f'{TREE_SHAP_BACKGROUND_SUPPORTED_SIZE} samples in the background dataset. '
f'A larger background dataset will be sampled with replacement to '
f'{TREE_SHAP_BACKGROUND_SUPPORTED_SIZE} instances.')

perturbation = 'interventional' if background_data is not None else 'tree_path_dependent'
self.background_data = background_data
self._explainer = shap.TreeExplainer(
Expand Down