def filter(self):
"""
The filter method is an important method which allows to summarize the local explainability
by using the user defined mask_params parameters which correspond to its use case.
"""
mask = [init_mask(self.summary['contrib_sorted'], True)]
if self.mask_params["features_to_hide"] is not None:
mask.append(
hide_contributions(self.summary['var_dict'],
features_list=self.check_features_name(
self.mask_params["features_to_hide"])))
if self.mask_params["threshold"] is not None:
mask.append(
cap_contributions(self.summary['contrib_sorted'],
threshold=self.mask_params["threshold"]))
if self.mask_params["positive"] is not None:
mask.append(
sign_contributions(self.summary['contrib_sorted'],
positive=self.mask_params["positive"]))
self.mask = combine_masks(mask)
if self.mask_params["max_contrib"] is not None:
self.mask = cutoff_contributions(mask=self.mask,
k=self.mask_params["max_contrib"])
self.masked_contributions = compute_masked_contributions(
self.summary['contrib_sorted'], self.mask)
def test_cutoff_contributions_3(self):
"""
Unit test cutoff contributions 3
"""
dataframe = pd.DataFrame({
'val1': [False, False, False],
'val2': [False, False, True],
'val3': [False, True, False],
'val4': [False, False, True]
})
output = cutoff_contributions(dataframe)
expected = pd.DataFrame({
'val1': [False, False, False],
'val2': [False, False, True],
'val3': [False, True, False],
'val4': [False, False, True]
})
pd.testing.assert_frame_equal(output, expected)
def cutoff_contributions(self, dataframe, max_contrib):
"""
The function cutoff_contributions computes a mask on a sorted contribution matrix.
It outputs True everywhere the contribution is in the top-k,
k being defined as an option by the user.
Parameters
----------
dataframe : pd.Dataframe
DataFrame is a sorted shapley matrix.
max_contrib: int
The k most important contributions to keep.
Returns
-------
pd.Dataframe
Mask indicating where contributions should be considered.
"""
return cutoff_contributions(dataframe, max_contrib)