Skip to content

Commit 30e18a4

Browse files
interpret-mlinterpret-ml
authored
Merge pull request #50 from wamartin-aml/wamartin-aml/pdpEbmContracts
Add pdp and ebm contracts
2 parents 8d3910d + dfae1d4 commit 30e18a4

File tree

2 files changed

+102
-18
lines changed

2 files changed

+102
-18
lines changed

‎python/interpret/blackbox/partialdependence.py‎

Lines changed: 27 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -79,7 +79,7 @@ def _gen_pdp(
7979
X_mut[:, col_idx] = grid_point
8080
ice_lines[:, idx] = predict_fn(X_mut)
8181
mean = np.mean(ice_lines, axis=0)
82-
# std = np.std(ice_lines, axis=0)
82+
std = np.std(ice_lines, axis=0)
8383

8484
ice_lines = ice_lines[
8585
np.random.choice(ice_lines.shape[0], num_ice_samples, replace=False), :
@@ -92,8 +92,8 @@ def _gen_pdp(
9292
"values": X[:, col_idx],
9393
"density": {"names": values, "scores": counts},
9494
# NOTE: We can take either bounds or background values, picked one.
95-
# 'upper_bounds': mean + std * std_coef,
96-
# 'lower_bounds': mean - std * std_coef,
95+
"upper_bounds": mean + std * std_coef,
96+
"lower_bounds": mean - std * std_coef,
9797
"background_scores": ice_lines,
9898
}
9999

@@ -102,6 +102,8 @@ def explain_global(self, name=None):
102102
name = gen_name_from_class(self)
103103

104104
data_dicts = []
105+
feature_list = []
106+
density_list = []
105107
for col_idx, feature in enumerate(self.feature_names):
106108
feature_type = self.feature_types[col_idx]
107109
pdp = PartialDependence._gen_pdp(
@@ -112,9 +114,30 @@ def explain_global(self, name=None):
112114
num_points=self.num_points,
113115
std_coef=self.std_coef,
114116
)
117+
feature_dict = {
118+
"feature_values": pdp["values"],
119+
"scores": pdp["scores"],
120+
"upper_bounds": pdp["upper_bounds"],
121+
"lower_bounds": pdp["lower_bounds"]
122+
}
123+
feature_list.append(feature_dict)
124+
density_list.append(pdp["density"])
115125
data_dicts.append(pdp)
116126

117-
internal_obj = {"overall": None, "specific": data_dicts}
127+
internal_obj = {"overall": None, "specific": data_dicts, "mli": [
128+
{
129+
"explanation_type": "pdp",
130+
"value": {
131+
"feature_list": feature_list
132+
}
133+
},
134+
{
135+
"explanation_type": "density",
136+
"value": {
137+
"density": density_list
138+
}
139+
}
140+
]}
118141

119142
selector = gen_global_selector(
120143
self.data, self.feature_names, self.feature_types, None

‎python/interpret/glassbox/ebm/ebm.py‎

Lines changed: 75 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -1026,6 +1026,8 @@ def explain_global(self, name=None):
10261026

10271027
# Add per feature graph
10281028
data_dicts = []
1029+
feature_list = []
1030+
density_list = []
10291031
for attribute_set_index, attribute_set in enumerate(self.attribute_sets_):
10301032
model_graph = self.attribute_set_models_[attribute_set_index]
10311033

@@ -1038,21 +1040,37 @@ def explain_global(self, name=None):
10381040
# bin_counts = self.preprocessor_.get_bin_counts(
10391041
# attribute_indexes[0]
10401042
# )
1043+
scores = list(model_graph)
1044+
upper_bounds = list(model_graph + errors)
1045+
lower_bounds = list(model_graph - errors)
1046+
density_dict = {
1047+
"names": self.preprocessor_.get_hist_edges(
1048+
attribute_indexes[0]
1049+
),
1050+
"scores": self.preprocessor_.get_hist_counts(
1051+
attribute_indexes[0]
1052+
),
1053+
}
1054+
1055+
feature_dict = {
1056+
"type": "univariate",
1057+
"names": bin_labels,
1058+
"scores": scores,
1059+
"scores_range": bounds,
1060+
"upper_bounds": upper_bounds,
1061+
"lower_bounds": lower_bounds,
1062+
}
1063+
feature_list.append(feature_dict)
1064+
density_list.append(density_dict)
1065+
10411066
data_dict = {
10421067
"type": "univariate",
10431068
"names": bin_labels,
1044-
"scores": list(model_graph),
1069+
"scores": scores,
10451070
"scores_range": bounds,
1046-
"upper_bounds": list(model_graph + errors),
1047-
"lower_bounds": list(model_graph - errors),
1048-
"density": {
1049-
"names": self.preprocessor_.get_hist_edges(
1050-
attribute_indexes[0]
1051-
),
1052-
"scores": self.preprocessor_.get_hist_counts(
1053-
attribute_indexes[0]
1054-
),
1055-
},
1071+
"upper_bounds": upper_bounds,
1072+
"lower_bounds": lower_bounds,
1073+
"density": density_dict,
10561074
}
10571075
data_dicts.append(data_dict)
10581076
elif len(attribute_indexes) == 2:
@@ -1062,6 +1080,17 @@ def explain_global(self, name=None):
10621080
bin_labels_right = self.preprocessor_.get_bin_labels(
10631081
attribute_indexes[1]
10641082
)
1083+
1084+
feature_dict = {
1085+
"type": "pairwise",
1086+
"left_names": bin_labels_left,
1087+
"right_names": bin_labels_right,
1088+
"scores": model_graph,
1089+
"scores_range": bounds,
1090+
}
1091+
feature_list.append(feature_dict)
1092+
density_list.append({})
1093+
10651094
data_dict = {
10661095
"type": "pairwise",
10671096
"left_names": bin_labels_left,
@@ -1078,7 +1107,20 @@ def explain_global(self, name=None):
10781107
"names": self.feature_names,
10791108
"scores": self.mean_abs_scores_,
10801109
}
1081-
internal_obj = {"overall": overall_dict, "specific": data_dicts}
1110+
internal_obj = {"overall": overall_dict, "specific": data_dicts, "mli": [
1111+
{
1112+
"explanation_type": "ebm_global",
1113+
"value": {
1114+
"feature_list": feature_list
1115+
}
1116+
},
1117+
{
1118+
"explanation_type": "density",
1119+
"value": {
1120+
"density": density_list
1121+
}
1122+
}
1123+
]}
10821124

10831125
return EBMExplanation(
10841126
"global",
@@ -1134,12 +1176,31 @@ def explain_local(self, X, y=None, name=None):
11341176
else:
11351177
scores = EBMUtils.regressor_predict(instances, self)
11361178

1179+
perf_list = []
11371180
for row_idx in range(n_rows):
1138-
data_dicts[row_idx]["perf"] = perf_dict(y, scores, row_idx)
1181+
perf = perf_dict(y, scores, row_idx)
1182+
perf_list.append(perf)
1183+
data_dicts[row_idx]["perf"] = perf
11391184

11401185
selector = gen_local_selector(instances, y, scores)
11411186

1142-
internal_obj = {"overall": None, "specific": data_dicts}
1187+
internal_obj = {"overall": None, "specific": data_dicts, "mli": [
1188+
{
1189+
"explanation_type": "ebm_local",
1190+
"value": {
1191+
"scores": self.attribute_set_models_,
1192+
"intercept": self.intercept_,
1193+
"perf": perf_list,
1194+
},
1195+
}
1196+
],
1197+
}
1198+
internal_obj["mli"].append(
1199+
{
1200+
"explanation_type": "evaluation_dataset",
1201+
"value": {"dataset_x": X, "dataset_y": y},
1202+
}
1203+
)
11431204

11441205
return EBMExplanation(
11451206
"local",

0 commit comments

Comments
 (0)