Merge pull request #5159 from PrimozGodec/fix-calibration

[FIX] Calibration model: Work with numpy data

Author: janezd

Orange/base.py

@@ -136,7 +136,11 @@ def __call__(self, data, progress_callback=None):
         progress_callback(0.1, "Fitting...")
         model = self._fit_model(data)
         model.used_vals = [np.unique(y).astype(int) for y in data.Y[:, None].T]
-        model.domain = data.domain
+        if not hasattr(model, "domain") or model.domain is None:
+            # some models set domain themself and it should be respected
+            # e.g. calibration learners set the base_learner's domain which
+            # would be wrongly overwritten if we set it here for any model
+            model.domain = data.domain
         model.supports_multiclass = self.supports_multiclass
         model.name = self.name
         model.original_domain = origdomain

Orange/classification/tests/test_calibration.py

@@ -4,6 +4,7 @@
 import numpy as np
 
 from Orange.base import Model
+from Orange.classification import LogisticRegressionLearner
 from Orange.classification.calibration import \
     ThresholdLearner, ThresholdClassifier, \
     CalibratedLearner, CalibratedClassifier
@@ -65,6 +66,18 @@ def test_non_binary_base(self):
         base_model.domain.class_var.is_discrete = False
         self.assertRaises(ValueError, ThresholdClassifier, base_model, 0.5)
 
+    def test_np_data(self):
+        """
+        Test ThresholdModel with numpy data.
+        When passing numpy data to model they should be already
+        transformed to models domain since model do not know how to do it.
+        """
+        data = Table('heart_disease')
+        base_learner = LogisticRegressionLearner()
+        model = ThresholdLearner(base_learner)(data)
+        res = model(model.data_to_model_domain(data).X)
+        self.assertTupleEqual((len(data),), res.shape)
+
 
 class TestThresholdLearner(unittest.TestCase):
     @patch("Orange.evaluation.performance_curves.Curves.from_results")
@@ -169,6 +182,18 @@ def test_calibrated_probs(self):
         calprobs = self.model.calibrated_probs(self.probs)
         np.testing.assert_almost_equal(calprobs, expprobs)
 
+    def test_np_data(self):
+        """
+        Test CalibratedClassifier with numpy data.
+        When passing numpy data to model they should be already
+        transformed to models domain since model do not know how to do it.
+        """
+        data = Table('heart_disease')
+        base_learner = LogisticRegressionLearner()
+        model = CalibratedLearner(base_learner)(data)
+        res = model(model.data_to_model_domain(data).X)
+        self.assertTupleEqual((len(data),), res.shape)
+
 
 class TestCalibratedLearner(unittest.TestCase):
     @patch("Orange.classification.calibration._SigmoidCalibration.fit")
@@ -207,3 +232,7 @@ def test_fit_storage(self, test_on_training, sigmoid_fit):
         for call, cls_probs in zip(sigmoid_fit.call_args_list,
                                    res.probabilities[0].T):
             np.testing.assert_equal(call[0][0], cls_probs)
+
+
+if __name__ == "__main__":
+    unittest.main()

Orange/tests/test_classification.py

@@ -245,20 +245,29 @@ def test_result_shape_numpy(self):
         Test whether results shapes are correct when testing on numpy data
         """
         iris = Table('iris')
+        iris_bin = Table(
+            Domain(
+                iris.domain.attributes,
+                DiscreteVariable("iris", values=["a", "b"])
+            ),
+            iris.X[:100], iris.Y[:100]
+        )
         for learner in all_learners():
             with self.subTest(learner.__name__):
-                try:
-                    model = learner()(iris)
-                except TypeError:
-                    # cannot be tested with default parameters
-                    continue
-                transformed_iris = model.data_to_model_domain(iris)
+                args = []
+                if learner in (ThresholdLearner, CalibratedLearner):
+                    args = [LogisticRegressionLearner()]
+                data = iris_bin if learner is ThresholdLearner else iris
+                model = learner(*args)(data)
+                transformed_iris = model.data_to_model_domain(data)
 
                 res = model(transformed_iris.X[0:5])
                 self.assertTupleEqual((5,), res.shape)
 
                 res = model(transformed_iris.X[0:1], model.Probs)
-                self.assertTupleEqual((1, 3), res.shape)
+                self.assertTupleEqual(
+                    (1, len(data.domain.class_var.values)), res.shape
+                )
 
 
 class ExpandProbabilitiesTest(unittest.TestCase):