OWAdaBoost: Use widget error messages

The widget simply crashed if an invalid learner was set as its base
estimator. Now, we use the widget Error messages and fail gracefully.

Author: pavlin-policar

Orange/widgets/classify/owadaboost.py

@@ -1,12 +1,13 @@
 from AnyQt.QtCore import Qt
 
+from Orange.classification import SklTreeLearner
 from Orange.classification.base_classification import LearnerClassification
 from Orange.data import Table
-from Orange.classification import SklTreeLearner
 from Orange.ensembles import SklAdaBoostLearner
 from Orange.widgets import gui
 from Orange.widgets.settings import Setting
 from Orange.widgets.utils.owlearnerwidget import OWBaseLearner
+from Orange.widgets.widget import Msg
 
 
 class OWAdaBoostClassification(OWBaseLearner):
@@ -28,6 +29,9 @@ class OWAdaBoostClassification(OWBaseLearner):
 
     DEFAULT_BASE_ESTIMATOR = SklTreeLearner()
 
+    class Error(OWBaseLearner.Error):
+        no_weight_support = Msg('The base learner does not support weights.')
+
     def add_main_layout(self):
         box = gui.widgetBox(self.controlArea, "Parameters")
         self.base_estimator = self.DEFAULT_BASE_ESTIMATOR
@@ -36,10 +40,11 @@ def add_main_layout(self):
 
         self.n_estimators_spin = gui.spin(
             box, self, "n_estimators", 1, 100, label="Number of estimators:",
-            alignment=Qt.AlignRight, callback=self.settings_changed)
+            alignment=Qt.AlignRight, controlWidth=80,
+            callback=self.settings_changed)
         self.learning_rate_spin = gui.doubleSpin(
             box, self, "learning_rate", 1e-5, 1.0, 1e-5, label="Learning rate:",
-            decimals=5, alignment=Qt.AlignRight, controlWidth=90,
+            decimals=5, alignment=Qt.AlignRight, controlWidth=80,
             callback=self.settings_changed)
         self.add_specific_parameters(box)
 
@@ -49,18 +54,25 @@ def add_specific_parameters(self, box):
             orientation=Qt.Horizontal, callback=self.settings_changed)
 
     def create_learner(self):
+        if self.base_estimator is None:
+            return None
         return self.LEARNER(
             base_estimator=self.base_estimator,
             n_estimators=self.n_estimators,
             learning_rate=self.learning_rate,
             preprocessors=self.preprocessors,
-            algorithm=self.losses[self.algorithm]
-        )
+            algorithm=self.losses[self.algorithm])
 
     def set_base_learner(self, learner):
-        self.base_estimator = learner if learner \
-            else self.DEFAULT_BASE_ESTIMATOR
-        self.base_label.setText("Base estimator: " + self.base_estimator.name)
+        self.Error.no_weight_support.clear()
+        if learner and not learner.supports_weights:
+            # Clear the error and reset to default base learner
+            self.Error.no_weight_support()
+            self.base_estimator = None
+            self.base_label.setText("Base estimator: INVALID")
+        else:
+            self.base_estimator = learner or self.DEFAULT_BASE_ESTIMATOR
+            self.base_label.setText("Base estimator: " + self.base_estimator.name)
         if self.auto_apply:
             self.apply()
 
@@ -76,7 +88,7 @@ def get_learner_parameters(self):
 
     a = QApplication(sys.argv)
     ow = OWAdaBoostClassification()
-    ow.set_data(Table("iris"))
+    ow.set_data(Table(sys.argv[1] if len(sys.argv) > 1 else 'iris'))
     ow.show()
     a.exec_()
     ow.saveSettings()

Orange/widgets/classify/tests/test_owadaboostclassification.py

@@ -1,9 +1,12 @@
 # Test methods with long descriptive names can omit docstrings
 # pylint: disable=missing-docstring
-from Orange.classification import SklTreeLearner, KNNLearner
+from Orange.classification import (
+    KNNLearner, RandomForestLearner, SklTreeLearner
+)
 from Orange.widgets.classify.owadaboost import OWAdaBoostClassification
-from Orange.widgets.tests.base import (WidgetTest, WidgetLearnerTestMixin,
-                                       ParameterMapping)
+from Orange.widgets.tests.base import (
+    WidgetTest, WidgetLearnerTestMixin, ParameterMapping
+)
 
 
 class TestOWAdaBoostClassification(WidgetTest, WidgetLearnerTestMixin):
@@ -30,10 +33,32 @@ def test_input_learner(self):
         output_base_est = self.get_output("Learner").params.get("base_estimator")
         self.assertEqual(output_base_est.max_depth, max_depth)
 
+    def test_input_learner_that_does_not_support_sample_weights(self):
+        self.send_signal("Learner", KNNLearner())
+        self.assertNotIsInstance(self.widget.base_estimator, KNNLearner)
+        self.assertIsNone(self.widget.base_estimator)
+        self.assertTrue(self.widget.Error.no_weight_support.is_shown())
+
+    def test_error_message_cleared_when_valid_learner_on_input(self):
+        # Disconnecting an invalid learner should use the default one and hide
+        # the error
+        self.send_signal("Learner", KNNLearner())
+        self.send_signal('Learner', None)
+        self.assertFalse(
+            self.widget.Error.no_weight_support.is_shown(),
+            'Error message was not hidden on input disconnect')
+        # Connecting a valid learner should also reset the error message
+        self.send_signal("Learner", KNNLearner())
+        self.send_signal('Learner', RandomForestLearner())
+        self.assertFalse(
+            self.widget.Error.no_weight_support.is_shown(),
+            'Error message was not hidden when a valid learner appeared on '
+            'input')
+
     def test_input_learner_disconnect(self):
         """Check base learner after disconnecting learner on the input"""
-        self.send_signal("Learner", KNNLearner())
-        self.assertIsInstance(self.widget.base_estimator, KNNLearner)
+        self.send_signal("Learner", RandomForestLearner())
+        self.assertIsInstance(self.widget.base_estimator, RandomForestLearner)
         self.send_signal("Learner", None)
         self.assertEqual(self.widget.base_estimator,
                          self.widget.DEFAULT_BASE_ESTIMATOR)

Orange/widgets/regression/owadaboostregression.py

@@ -1,9 +1,9 @@
 from AnyQt.QtCore import Qt
 
-from Orange.regression.base_regression import LearnerRegression
-from Orange.regression import SklTreeRegressionLearner
 from Orange.data import Table
 from Orange.ensembles import SklAdaBoostRegressionLearner
+from Orange.regression import SklTreeRegressionLearner
+from Orange.regression.base_regression import LearnerRegression
 from Orange.widgets import gui
 from Orange.widgets.classify import owadaboost
 from Orange.widgets.settings import Setting
@@ -36,8 +36,7 @@ def create_learner(self):
             n_estimators=self.n_estimators,
             learning_rate=self.learning_rate,
             preprocessors=self.preprocessors,
-            loss=self.losses[self.loss].lower()
-        )
+            loss=self.losses[self.loss].lower())
 
     def get_learner_parameters(self):
         return (("Base estimator", self.base_estimator),
@@ -51,7 +50,7 @@ def get_learner_parameters(self):
 
     a = QApplication(sys.argv)
     ow = OWAdaBoostRegression()
-    ow.set_data(Table("housing"))
+    ow.set_data(Table(sys.argv[1] if len(sys.argv) > 1 else 'housing'))
     ow.show()
     a.exec_()
     ow.saveSettings()

Orange/widgets/regression/tests/test_owadaboostregression.py

@@ -1,9 +1,14 @@
 # Test methods with long descriptive names can omit docstrings
 # pylint: disable=missing-docstring
-from Orange.regression import SklTreeRegressionLearner, KNNRegressionLearner
+from Orange.regression import (
+    SklTreeRegressionLearner,
+    KNNRegressionLearner,
+    RandomForestRegressionLearner
+)
 from Orange.widgets.regression.owadaboostregression import OWAdaBoostRegression
-from Orange.widgets.tests.base import (WidgetTest, WidgetLearnerTestMixin,
-                                       ParameterMapping)
+from Orange.widgets.tests.base import (
+    WidgetTest, WidgetLearnerTestMixin, ParameterMapping
+)
 
 
 class TestOWAdaBoostRegression(WidgetTest, WidgetLearnerTestMixin):
@@ -31,10 +36,34 @@ def test_input_learner(self):
         output_base_est = self.get_output("Learner").params.get("base_estimator")
         self.assertEqual(output_base_est.max_depth, max_depth)
 
+    def test_input_learner_that_does_not_support_sample_weights(self):
+        self.send_signal("Learner", KNNRegressionLearner())
+        self.assertNotIsInstance(
+            self.widget.base_estimator, KNNRegressionLearner)
+        self.assertIsNone(self.widget.base_estimator)
+        self.assertTrue(self.widget.Error.no_weight_support.is_shown())
+
+    def test_error_message_cleared_when_valid_learner_on_input(self):
+        # Disconnecting an invalid learner should use the default one and hide
+        # the error
+        self.send_signal("Learner", KNNRegressionLearner())
+        self.send_signal('Learner', None)
+        self.assertFalse(
+            self.widget.Error.no_weight_support.is_shown(),
+            'Error message was not hidden on input disconnect')
+        # Connecting a valid learner should also reset the error message
+        self.send_signal("Learner", KNNRegressionLearner())
+        self.send_signal('Learner', RandomForestRegressionLearner())
+        self.assertFalse(
+            self.widget.Error.no_weight_support.is_shown(),
+            'Error message was not hidden when a valid learner appeared on '
+            'input')
+
     def test_input_learner_disconnect(self):
         """Check base learner after disconnecting learner on the input"""
-        self.send_signal("Learner", KNNRegressionLearner())
-        self.assertIsInstance(self.widget.base_estimator, KNNRegressionLearner)
+        self.send_signal("Learner", RandomForestRegressionLearner())
+        self.assertIsInstance(
+            self.widget.base_estimator, RandomForestRegressionLearner)
         self.send_signal("Learner", None)
         self.assertEqual(self.widget.base_estimator,
                          self.widget.DEFAULT_BASE_ESTIMATOR)

Orange/widgets/utils/owlearnerwidget.py

@@ -165,7 +165,8 @@ def apply(self):
 
     def update_learner(self):
         self.learner = self.create_learner()
-        self.learner.name = self.learner_name
+        if self.learner is not None:
+            self.learner.name = self.learner_name
         self.send("Learner", self.learner)
         self.outdated_settings = False
         self.Warning.outdated_learner.clear()