ICE: Handle learner that needs discretization

orangecontrib/explain/inspection.py

@@ -189,13 +189,7 @@ def individual_condition_expectation(
     _check_data(data)
 
     # implicit check if feature in data.domain
-    needs_pp = _check_model(model, data)
-
-    # values should not be preprocessed
-    orig_values = data[:, feature].X.flatten()
-    _, index = np.unique(orig_values, return_index=True)
-    orig_values = orig_values[index]
-    if needs_pp:
+    if _check_model(model, data):
         data = model.data_to_model_domain(data)
 
     assert feature.name in [a.name for a in data.domain.attributes]
@@ -209,7 +203,7 @@ def dummy_fit(*_, **__):
     model.fit_ = dummy_fit
     if model.domain.class_var.is_discrete:
         model._estimator_type = "classifier"
-        model.classes_ = model.domain.class_var.values
+        model.classes_ = np.array(model.domain.class_var.values)
     else:
         model._estimator_type = "regressor"
 
@@ -221,7 +215,7 @@ def dummy_fit(*_, **__):
                              grid_resolution=grid_resolution,
                              kind=kind)
 
-    results = {"average": dep["average"], "values": orig_values}
+    results = {"average": dep["average"], "values": dep["grid_values"][0]}
     if kind == "both":
         results["individual"] = dep["individual"]
 

orangecontrib/explain/tests/test_explainer.py

@@ -2,7 +2,6 @@
 import unittest
 
 import numpy as np
-from numpy.testing import assert_array_equal
 from Orange.classification import (
     LogisticRegressionLearner,
     RandomForestLearner,
@@ -11,6 +10,10 @@
     TreeLearner,
     ThresholdLearner,
 )
+try:
+    from Orange.classification import ScoringSheetLearner
+except ImportError:
+    ScoringSheetLearner = None
 from Orange.data import Table, Domain, ContinuousVariable
 from Orange.regression import LinearRegressionLearner, CurveFitLearner
 from Orange.tests import test_regression, test_classification
@@ -206,7 +209,7 @@ def test_gradient_boosting_shape(self):
         )
         self.assertEqual(len(shap_values), 2)
         self.assertEqual(len(base_value), 2)
-        assert_array_equal(-shap_values[0], shap_values[1])
+        np.testing.assert_array_almost_equal(-shap_values[0], shap_values[1])
 
         learner = GBLearner()
         model = learner(self.hearth_disease)
@@ -215,7 +218,7 @@ def test_gradient_boosting_shape(self):
         )
         self.assertEqual(len(shap_values), 2)
         self.assertEqual(len(base_value), 2)
-        assert_array_equal(-shap_values[0], shap_values[1])
+        np.testing.assert_array_almost_equal(-shap_values[0], shap_values[1])
 
         learner = XGBRFLearner()
         model = learner(self.hearth_disease)
@@ -224,7 +227,7 @@ def test_gradient_boosting_shape(self):
         )
         self.assertEqual(len(shap_values), 2)
         self.assertEqual(len(base_value), 2)
-        assert_array_equal(-shap_values[0], shap_values[1])
+        np.testing.assert_array_almost_equal(-shap_values[0], shap_values[1])
 
     @unittest.skipIf(XGBLearner is None, "Missing 'xgboost' package")
     def test_remove_workaround(self):
@@ -253,8 +256,8 @@ def test_all_classifiers(self):
         """ Test explanation for all classifiers """
         for learner in test_classification.all_learners():
             with self.subTest(learner):
-                if learner == ThresholdLearner:
-                    # ThresholdLearner require binary class
+                if learner in (ThresholdLearner, ScoringSheetLearner):
+                    # ThresholdLearner and ScoringSheetLearner require binary class
                     continue
                 kwargs = {}
                 if "base_learner" in inspect.signature(learner).parameters:

orangecontrib/explain/tests/test_inspection.py

@@ -319,17 +319,6 @@ def test_continuous_class(self):
         self.assertEqual(res["individual"].shape, (1, 506, 504))
         self.assertEqual(res["values"].shape, (504,))
 
-    def test_retain_original_values(self):
-        data = self.housing
-        nn = NNRegressionLearner(random_state=0)(data)
-        res_nn = individual_condition_expectation(nn, data, data.domain[0])
-        rf = RandomForestRegressionLearner(n_estimators=10, random_state=0)(data)
-        res_rf = individual_condition_expectation(rf, data, data.domain[0])
-        self.assertEqual(res_nn["values"].min(), res_rf["values"].min())
-        self.assertEqual(res_nn["values"].max(), res_rf["values"].max())
-        self.assertEqual(res_nn["values"].shape, res_rf["values"].shape)
-        self.assertEqual(len(set(res_nn["values"])), len(res_rf["values"]))
-
     def test_multi_class(self):
         data = self.iris
         model = RandomForestLearner(n_estimators=10, random_state=0)(data)

orangecontrib/explain/widgets/owice.py

@@ -508,6 +508,9 @@ class Error(OWWidget.Error):
         not_enough_data = Msg("At least two instances are needed.")
         no_cont_features = Msg("At least one numeric feature is required.")
 
+    class Warning(OWWidget.Warning):
+        pp_feature = Msg("Selected feature has been preprocessed.")
+
     class Information(OWWidget.Information):
         data_sampled = Msg("Data has been sampled.")
 
@@ -762,6 +765,7 @@ def clear(self):
         self.cancel()
         self.Error.domain_transform_err.clear()
         self.Error.unknown_err.clear()
+        self.Warning.pp_feature.clear()
         self.graph.clear_all()
 
     def setup_plot(self):
@@ -773,6 +777,13 @@ def setup_plot(self):
         y_average = self.__results.y_average[self.target_index]
         y_individual = self.__results.y_individual[self.target_index]
 
+        data = self.data[self.__sampled_mask]
+        orig_values = data[:, self.feature].X.flatten()
+        _, index = np.unique(orig_values, return_index=True)
+        orig_values = orig_values[index]
+        if len(orig_values) != len(x_data) or (orig_values != x_data).any():
+            self.Warning.pp_feature()
+
         class_var: Variable = self.model.original_domain.class_var
         if class_var.is_discrete:
             cls_val = class_var.values[self.target_index]

orangecontrib/explain/widgets/tests/test_owice.py

@@ -5,7 +5,8 @@
 from AnyQt.QtCore import Qt, QPointF
 
 from Orange.classification import RandomForestLearner, CalibratedLearner, \
-    ThresholdLearner, SimpleRandomForestLearner as SimpleRandomForestClassifier
+    ThresholdLearner, NaiveBayesLearner, \
+    SimpleRandomForestLearner as SimpleRandomForestClassifier
 from Orange.data import Table
 from Orange.data.table import DomainTransformationError
 from Orange.regression import RandomForestRegressionLearner, \
@@ -177,6 +178,18 @@ def test_send_report(self):
         self.send_signal(self.widget.Inputs.model, self.rf_reg)
         self.widget.send_report()
 
+    def test_naive_bayes(self):
+        data = self.iris
+        self.send_signal(self.widget.Inputs.data, data)
+
+        self.send_signal(self.widget.Inputs.model, NaiveBayesLearner()(data))
+        self.wait_until_finished()
+        self.assertTrue(self.widget.Warning.pp_feature.is_shown())
+
+        self.send_signal(self.widget.Inputs.model, self.rf_cls)
+        self.wait_until_finished()
+        self.assertFalse(self.widget.Warning.pp_feature.is_shown())
+
 
 if __name__ == "__main__":
     unittest.main()

tox.ini

@@ -28,6 +28,7 @@ deps =
     oldest: pandas==1.4.0
     oldest: scikit-learn==1.3.0
     oldest: scipy==1.9.0
+    oldest: xgboost==2.0.0
     latest: https://github.com/biolab/orange3/archive/refs/heads/master.zip#egg=orange3
     latest: https://github.com/biolab/orange-canvas-core/archive/refs/heads/master.zip#egg=orange-canvas-core
     latest: https://github.com/biolab/orange-widget-base/archive/refs/heads/master.zip#egg=orange-widget-base