def _make_single_prediction_shap_table(pipeline, input_features, top_k=3, training_data=None,
include_shap_values=False, output_format="text"):
"""Creates table summarizing the top_k positive and top_k negative contributing features to the prediction of a single datapoint.
Arguments:
pipeline (PipelineBase): Fitted pipeline whose predictions we want to explain with SHAP.
input_features (pd.DataFrame): Dataframe of features - needs to correspond to data the pipeline was fit on.
top_k (int): How many of the highest/lowest features to include in the table.
training_data (pd.DataFrame): Training data the pipeline was fit on.
This is required for non-tree estimators because we need a sample of training data for the KernelSHAP algorithm.
include_shap_values (bool): Whether the SHAP values should be included in an extra column in the output.
Default is False.
Returns:
str: Table
"""
pipeline_features = pipeline.compute_estimator_features(input_features)
shap_values = _compute_shap_values(pipeline, pipeline_features, training_data)
normalized_shap_values = _normalize_shap_values(shap_values)
class_names = None
if hasattr(pipeline, "classes_"):
class_names = pipeline.classes_
table_makers = {ProblemTypes.REGRESSION: _RegressionSHAPTable(),
ProblemTypes.BINARY: _BinarySHAPTable(class_names),
ProblemTypes.MULTICLASS: _MultiClassSHAPTable(class_names)}
table_maker_class = table_makers[pipeline.problem_type]
table_maker = table_maker_class.make_text if output_format == "text" else table_maker_class.make_dict
return table_maker(shap_values, normalized_shap_values, pipeline_features, top_k, include_shap_values)
def _make_single_prediction_shap_table(pipeline,
input_features,
y,
index_to_explain,
top_k=3,
include_shap_values=False,
output_format="text"):
"""Creates table summarizing the top_k_features positive and top_k_features negative contributing features to the prediction of a single datapoint.
Arguments:
pipeline (PipelineBase): Fitted pipeline whose predictions we want to explain with SHAP.
input_features (pd.DataFrame): Dataframe of features - needs to correspond to data the pipeline was fit on.
top_k (int): How many of the highest/lowest features to include in the table.
training_data (pd.DataFrame): Training data the pipeline was fit on.
This is required for non-tree estimators because we need a sample of training data for the KernelSHAP algorithm.
include_shap_values (bool): Whether the SHAP values should be included in an extra column in the output.
Default is False.
output_format (str): The desired format of the output. Can be "text", "dict", or "dataframe".
Returns:
str: Table
Raises:
ValueError: if requested index results in a NaN in the computed features.
"""
pipeline_features = pipeline.compute_estimator_features(input_features,
y).to_dataframe()
pipeline_features_row = pipeline_features.iloc[[index_to_explain]]
if pipeline_features_row.isna().any(axis=None):
raise ValueError(
f"Requested index ({index_to_explain}) produces NaN in features.")
shap_values = _compute_shap_values(
pipeline,
pipeline_features_row,
training_data=pipeline_features.dropna(axis=0))
normalized_shap_values = _normalize_shap_values(shap_values)
class_names = None
if hasattr(pipeline, "classes_"):
class_names = pipeline.classes_
table_makers = {
ProblemTypes.REGRESSION: _RegressionSHAPTable(),
ProblemTypes.BINARY: _BinarySHAPTable(class_names),
ProblemTypes.MULTICLASS: _MultiClassSHAPTable(class_names),
ProblemTypes.TIME_SERIES_REGRESSION: _RegressionSHAPTable(),
ProblemTypes.TIME_SERIES_BINARY: _BinarySHAPTable(class_names),
ProblemTypes.TIME_SERIES_MULTICLASS: _MultiClassSHAPTable(class_names)
}
table_maker_class = table_makers[pipeline.problem_type]
table_maker = {
"text": table_maker_class.make_text,
"dict": table_maker_class.make_dict,
"dataframe": table_maker_class.make_dataframe
}[output_format]
return table_maker(shap_values, normalized_shap_values,
pipeline_features_row, top_k, include_shap_values)
def test_value_errors_raised(mock_tree_explainer, pipeline, exception, match):
if "xgboost" in pipeline.name.lower():
pytest.importorskip("xgboost", "Skipping test because xgboost is not installed.")
if "catboost" in pipeline.name.lower():
pytest.importorskip("catboost", "Skipping test because catboost is not installed.")
with pytest.raises(exception, match=match):
_ = _compute_shap_values(pipeline({"pipeline": {"gap": 1, "max_delay": 1}}), pd.DataFrame(np.random.random((2, 16))))
def test_compute_shap_values_catches_shap_tree_warnings(mock_tree_explainer, mock_debug, X_y_binary, caplog):
X, y = X_y_binary
pipeline = BinaryClassificationPipeline(["Random Forest Classifier"])
def raise_warning_from_shap(estimator, feature_perturbation):
warnings.warn("Shap raised a warning!")
mock = MagicMock()
mock.shap_values.return_value = np.zeros(10)
return mock
mock_tree_explainer.side_effect = raise_warning_from_shap
_ = _compute_shap_values(pipeline, pd.DataFrame(X))
mock_debug.debug.assert_called_with("_compute_shap_values TreeExplainer: Shap raised a warning!")
def calculate_shap_for_test(training_data, y, pipeline, n_points_to_explain):
"""Helper function to compute the SHAP values for n_points_to_explain for a given pipeline."""
points_to_explain = training_data[:n_points_to_explain]
pipeline.fit(training_data, y)
return _compute_shap_values(pipeline, pd.DataFrame(points_to_explain), training_data)