def test_surrogate_error_tree_int_categorical(self):
X_train, X_test, y_train, y_test, categorical_features = \
create_adult_census_data()
model = create_kneighbors_classifier(X_train, y_train)
run_error_analyzer(model, X_test, y_test, list(X_train.columns),
categorical_features)
def test_matrix_filter_adult_census_quantile_binning(
self, string_labels, metric):
(X_train, X_test, y_train, y_test,
categorical_features) = create_adult_census_data(string_labels)
model_task = ModelTask.CLASSIFICATION
feature_names = X_test.columns.tolist()
matrix_features = ['Capital Gain']
# validate quantile binning for column with many zero values
model = create_kneighbors_classifier(X_train, y_train)
# validate warning printed
err_capg = ("Removing duplicate bin edges for quantile binning of "
"feature Capital Gain. There are too many duplicate "
"values for the specified number of bins.")
with pytest.warns(UserWarning, match=err_capg):
run_error_analyzer(model,
X_test,
y_test,
feature_names,
categorical_features,
model_task=model_task,
matrix_features=matrix_features,
quantile_binning=True,
metric=metric)
matrix_features = ['Capital Gain', 'Capital Loss']
err_capl = ("Removing duplicate bin edges for quantile binning of "
"feature Capital Loss. There are too many duplicate "
"values for the specified number of bins.")
with pytest.warns(UserWarning) as warninfo:
run_error_analyzer(model,
X_test,
y_test,
feature_names,
categorical_features,
model_task=model_task,
matrix_features=matrix_features,
quantile_binning=True,
metric=metric)
warns = {(warn.category, warn.message.args[0]) for warn in warninfo}
expected = {(UserWarning, err_capg), (UserWarning, err_capl)}
for expected_warning in expected:
assert expected_warning in warns
def test_traverse_tree(self):
X_train, X_test, y_train, y_test, categorical_features = \
create_adult_census_data()
model = create_kneighbors_classifier(X_train, y_train)
feature_names = list(X_train.columns)
error_analyzer = ModelAnalyzer(model, X_test, y_test,
feature_names,
categorical_features)
categorical_info = get_categorical_info(error_analyzer,
feature_names)
cat_ind_reindexed, categories_reindexed = categorical_info
pred_y = model.predict(X_test)
diff = pred_y != y_test
max_depth = 3
num_leaves = 31
surrogate = create_surrogate_model(error_analyzer,
X_test,
diff,
max_depth,
num_leaves,
cat_ind_reindexed)
model_json = surrogate._Booster.dump_model()
tree_structure = model_json["tree_info"][0]['tree_structure']
max_split_index = get_max_split_index(tree_structure) + 1
filtered_indexed_df = X_test
filtered_indexed_df[DIFF] = diff
filtered_indexed_df[TRUE_Y] = y_test
filtered_indexed_df[PRED_Y] = pred_y
json_tree = traverse(filtered_indexed_df,
tree_structure,
max_split_index,
(categories_reindexed,
cat_ind_reindexed),
[],
feature_names,
metric=error_analyzer.metric)
# create dictionary from json tree id to values
json_tree_dict = {}
for entry in json_tree:
json_tree_dict[entry['id']] = entry
validate_traversed_tree(tree_structure, json_tree_dict,
max_split_index, feature_names)