sync tests

tests/explainability/test_anchors_extended.py

@@ -93,7 +93,7 @@ def test_anchors_fit_and_explain_coverage(model_and_data):
     explanation = anchorsExtendedExplainer.explain(
         explain_data.head(1).values, threshold=0.95
     )
-    assert explanation.data['coverage'] == pytest.approx(0.42, 0.01)
+    assert explanation.data['precision'] > 0.95
 
 def test_anchors_feature_importance_obtention(model_and_data):
     """

tests/explainability/test_serializer.py

@@ -1,15 +1,17 @@
 from mercury.explainability.explainers import (
     AnchorsWithImportanceExplainer,
+    ALEExplainer,
     MercuryExplainer
 )
 import pytest
 import pickle
 import pandas as pd
+import numpy as np
 import os
 
 
 @pytest.fixture(scope="session")
-def model_and_data():
+def model_and_data_anchors():
     logRegModel = pickle.load(open('tests/explainability/model_and_data/FICO_lr_model.pkl', 'rb'))
     fit_data = pd.read_csv('tests/explainability/model_and_data/fit_data_red.csv', index_col=0)
     explain_data = pd.read_csv('tests/explainability/model_and_data/explain_data.csv', index_col=0)
@@ -19,17 +21,25 @@ def model_and_data():
         'explain_data': explain_data
     }
 
+@pytest.fixture(scope='session')
+def model_and_data_ale():
+    logRegModel = pickle.load(open('./tests/explainability/model_and_data/FICO_lr_model.pkl', 'rb'))
+    fit_data = pd.read_csv('./tests/explainability/model_and_data_pyspark/data_ale.csv', index_col=0)
+    return {
+        'logRegModel': logRegModel,
+        'data': fit_data,
+    }
 
 pytestmark = pytest.mark.usefixtures("model_and_data")
 
-def test_serializer_explainer(model_and_data):
+def test_serializer_explainer(model_and_data_anchors):
     """
         Testing out that the explainers are properly saved and then loaded
         back.
     """
-    logRegModel = model_and_data['logRegModel']
-    fit_data = model_and_data['fit_data']
-    explain_data = model_and_data['explain_data']
+    logRegModel = model_and_data_anchors['logRegModel']
+    fit_data = model_and_data_anchors['fit_data']
+    explain_data = model_and_data_anchors['explain_data']
     feature_names = list(explain_data.columns)
 
     TEST_FILE = "/tmp/explainer.pkl"
@@ -46,4 +56,56 @@ def test_serializer_explainer(model_and_data):
     assert type(anchorsExtendedExplainer_recovered) ==\
          AnchorsWithImportanceExplainer, "Bad load"
 
+    os.remove(TEST_FILE)
+
+def test_serializer_anchors_with_importance_explainer(model_and_data_anchors):
+
+    logRegModel = model_and_data_anchors['logRegModel']
+    fit_data = model_and_data_anchors['fit_data']
+    explain_data = model_and_data_anchors['explain_data']
+    feature_names = list(explain_data.columns)
+
+    TEST_FILE = "/tmp/explainer_anchor.pkl"
+
+    explainer = AnchorsWithImportanceExplainer(
+        train_data=fit_data,
+        predict_fn=logRegModel.predict_proba,
+        feature_names=feature_names
+    )
+    explainer.save(TEST_FILE)
+
+    explainer_loaded = MercuryExplainer.load(TEST_FILE)
+    assert isinstance(explainer_loaded, AnchorsWithImportanceExplainer)
+    assert explainer.params == explainer_loaded.params
+    assert explainer.feature_values == explainer_loaded.feature_values
+
+    # We are able to execute explain
+    explanation_loaded = explainer_loaded.explain(fit_data.values[0])
+
+    os.remove(TEST_FILE)
+
+def test_serializer_ale_explainer(model_and_data_ale):
+
+    model = model_and_data_ale['logRegModel']
+    data_pd = model_and_data_ale['data']
+    features = [c for c in list(data_pd.columns) if c not in ['label']]
+
+    TEST_FILE = "/tmp/explainer_ale.pkl"
+
+    explainer = ALEExplainer(
+        lambda x: model.predict_proba(x),
+        target_names="label"
+    )
+    explainer.save(TEST_FILE)
+
+    explainer_loaded = MercuryExplainer.load(TEST_FILE)
+    isinstance(explainer_loaded, ALEExplainer)
+
+    # Check explanations
+    explanation = explainer.explain(data_pd[features])
+    explanation_loaded = explainer_loaded.explain(data_pd[features])
+
+    for i in range(len(explanation.data)):
+        assert np.all(explanation.data['ale_values'][i] == explanation_loaded.data['ale_values'][i])
+
     os.remove(TEST_FILE)