Merge pull request #5710 from janezd/baselearnerwidget-pp-warning

[ENH] Learner widgets: Inform about potential problems when overriding preprocessors

Author: lanzagar

Orange/widgets/model/owadaboost.py

@@ -102,8 +102,7 @@ def set_base_learner(self, learner):
             self.base_estimator = learner or self.DEFAULT_BASE_ESTIMATOR
             self.base_label.setText(
                 "Base estimator: %s" % self.base_estimator.name.title())
-        if self.auto_apply:
-            self.apply()
+        self.learner = self.model = None
 
     def get_learner_parameters(self):
         return (("Base estimator", self.base_estimator),

Orange/widgets/model/owcalibratedlearner.py

@@ -62,7 +62,7 @@ def add_main_layout(self):
     def set_learner(self, learner):
         self.base_learner = learner
         self._set_default_name()
-        self.unconditional_apply()
+        self.learner = self.model = None
 
     def _set_default_name(self):
 

Orange/widgets/model/owcurvefit.py

@@ -395,6 +395,7 @@ def __insert_into_expression(self, what: str, offset=0):
 
     def set_data(self, data: Optional[Table]):
         self.Warning.data_missing(shown=not bool(data))
+        self.learner = None
         super().set_data(data)
         self.__clear()
 
@@ -419,7 +420,7 @@ def handleNewSignals(self):
         self.__init_models()
         self.__enable_controls()
         self.__set_pending()
-        self.unconditional_apply()
+        super().handleNewSignals()
 
     def __preprocess_data(self):
         self.__pp_data = preprocess(self.data, self.preprocessors)

Orange/widgets/model/owlinearregression.py

@@ -94,9 +94,6 @@ def add_main_layout(self):
         self.controls.alpha_index.setEnabled(self.reg_type != self.OLS)
         self.l2_ratio_slider.setEnabled(self.reg_type == self.Elastic)
 
-    def handleNewSignals(self):
-        self.apply()
-
     def _intercept_changed(self):
         self.apply()
 

Orange/widgets/model/owstack.py

@@ -33,22 +33,26 @@ def add_main_layout(self):
     @Inputs.learners
     def set_learner(self, index: int, learner: Learner):
         self.learners[index] = learner
+        self._invalidate()
 
     @Inputs.learners.insert
     def insert_learner(self, index, learner):
         self.learners.insert(index, learner)
+        self._invalidate()
 
     @Inputs.learners.remove
     def remove_learner(self, index):
         self.learners.pop(index)
+        self._invalidate()
 
     @Inputs.aggregate
     def set_aggregate(self, aggregate):
         self.aggregate = aggregate
+        self._invalidate()
 
-    def handleNewSignals(self):
-        super().handleNewSignals()
-        self.apply()
+    def _invalidate(self):
+        self.learner = self.model = None
+        # ... and handleNewSignals will do the rest
 
     def create_learner(self):
         if not self.learners:

Orange/widgets/utils/owlearnerwidget.py

@@ -79,6 +79,14 @@ class Error(OWWidget.Error):
     class Warning(OWWidget.Warning):
         outdated_learner = Msg("Press Apply to submit changes.")
 
+    class Information(OWWidget.Information):
+        ignored_preprocessors = Msg(
+            "Ignoring default preprocessing.\n"
+            "Default preprocessing, such as scaling, one-hot encoding and "
+            "treatment of missing data, has been replaced with user-specified "
+            "preprocessors. Problems may occur if these are inadequate "
+            "for the given data.")
+
     class Inputs:
         data = Input("Data", Table)
         preprocessor = Input("Preprocessor", Preprocess)
@@ -90,6 +98,8 @@ class Outputs:
 
     OUTPUT_MODEL_NAME = Outputs.model.name  # Attr for backcompat w/ self.send() code
 
+    _SEND, _SOFT, _UPDATE = range(3)
+
     def __init__(self, preprocessors=None):
         super().__init__()
         self.__default_learner_name = ""
@@ -99,6 +109,7 @@ def __init__(self, preprocessors=None):
         self.model = None
         self.preprocessors = preprocessors
         self.outdated_settings = False
+        self.__apply_level = []
 
         self.setup_layout()
         QTimer.singleShot(0, getattr(self, "unconditional_apply", self.apply))
@@ -144,7 +155,8 @@ def set_default_learner_name(self, name: str) -> None:
     @Inputs.preprocessor
     def set_preprocessor(self, preprocessor):
         self.preprocessors = preprocessor
-        self.apply()
+        # invalidate learner and model, so handleNewSignals will renew them
+        self.learner = self.model = None
 
     @Inputs.data
     @check_sql_input
@@ -164,23 +176,50 @@ def set_data(self, data):
                     "Select one with the Select Columns widget.")
             self.data = None
 
-        self.update_model()
+        # invalidate the model so that handleNewSignals will update it
+        self.model = None
+
 
     def apply(self):
+        level, self.__apply_level = max(self.__apply_level, default=self._UPDATE), []
         """Applies learner and sends new model."""
-        self.update_learner()
-        self.update_model()
+        if level == self._SEND:
+            self._send_learner()
+            self._send_model()
+        elif level == self._UPDATE:
+            self.update_learner()
+            self.update_model()
+        else:
+            self.learner or self.update_learner()
+            self.model or self.update_model()
+
+    def apply_as(self, level, unconditional=False):
+        self.__apply_level.append(level)
+        if unconditional:
+            self.unconditional_apply()
+        else:
+            self.apply()
 
     def update_learner(self):
         self.learner = self.create_learner()
         if self.learner and issubclass(self.LEARNER, Fitter):
             self.learner.use_default_preprocessors = True
         if self.learner is not None:
             self.learner.name = self.effective_learner_name()
+        self._send_learner()
+
+    def _send_learner(self):
         self.Outputs.learner.send(self.learner)
         self.outdated_settings = False
         self.Warning.outdated_learner.clear()
 
+    def handleNewSignals(self):
+        self.apply_as(self._SOFT, True)
+        self.Information.ignored_preprocessors(
+            shown=not getattr(self.learner, "use_default_preprocessors", False)
+                  and getattr(self.LEARNER, "preprocessors", False)
+                  and self.preprocessors is not None)
+
     def show_fitting_failed(self, exc):
         """Show error when fitting fails.
             Derived widgets can override this to show more specific messages."""
@@ -197,6 +236,9 @@ def update_model(self):
             else:
                 self.model.name = self.learner_name or self.captionTitle
                 self.model.instances = self.data
+        self._send_model()
+
+    def _send_model(self):
         self.Outputs.model.send(self.model)
 
     def check_data(self):
@@ -223,15 +265,12 @@ def settings_changed(self, *args, **kwargs):
         self.Warning.outdated_learner(shown=not self.auto_apply)
         self.apply()
 
-    def _change_name(self, instance, output):
-        if instance:
-            instance.name = self.effective_learner_name()
-            if self.auto_apply:
-                output.send(instance)
-
     def learner_name_changed(self):
-        self._change_name(self.learner, self.Outputs.learner)
-        self._change_name(self.model, self.Outputs.model)
+        if self.model is not None:
+            self.model.name = self.effective_learner_name()
+        if self.learner is not None:
+            self.learner.name = self.effective_learner_name()
+        self.apply_as(self._SEND)
 
     def effective_learner_name(self):
         """Return the effective learner name."""
@@ -272,7 +311,6 @@ def add_main_layout(self):
         Override this method for laying out any learner-specific parameter controls.
         See setup_layout() method for execution order.
         """
-        pass
 
     def add_classification_layout(self, box):
         """Creates layout for classification specific options.
@@ -281,7 +319,6 @@ def add_classification_layout(self, box):
         and regression learners require different options.
         See `setup_layout()` method for execution order.
         """
-        pass
 
     def add_regression_layout(self, box):
         """Creates layout for regression specific options.
@@ -290,7 +327,6 @@ def add_regression_layout(self, box):
         and regression learners require different options.
         See `setup_layout()` method for execution order.
         """
-        pass
 
     def add_learner_name_widget(self):
         self.name_line_edit = gui.lineEdit(

Orange/widgets/utils/tests/test_owlearnerwidget.py

@@ -1,4 +1,4 @@
-from unittest.mock import Mock
+from unittest.mock import Mock, patch
 
 import scipy.sparse as sp
 
@@ -218,3 +218,73 @@ def check_name(name):
         check_name("Bar")
         w.set_default_learner_name("")
         check_name("Blarg")
+
+    def test_preprocessor_warning(self):
+        class TestLearnerNoPreprocess(Learner):
+            name = "Test"
+            __returns__ = Mock()
+
+        class TestWidgetNoPreprocess(OWBaseLearner):
+            name = "Test"
+            LEARNER = TestLearnerNoPreprocess
+
+        class TestLearnerPreprocess(Learner):
+            name = "Test"
+            preprocessors = [Mock()]
+            __returns__ = Mock()
+
+        class TestWidgetPreprocess(OWBaseLearner):
+            name = "Test"
+            LEARNER = TestLearnerPreprocess
+
+        class TestFitterPreprocess(Fitter):
+            name = "Test"
+            preprocessors = [Mock()]
+            __returns__ = Mock()
+
+        class TestWidgetPreprocessFit(OWBaseLearner):
+            name = "Test"
+            LEARNER = TestFitterPreprocess
+
+        wno = self.create_widget(TestWidgetNoPreprocess)
+        wyes = self.create_widget(TestWidgetPreprocess)
+        wfit = self.create_widget(TestWidgetPreprocessFit)
+
+        self.assertFalse(wno.Information.ignored_preprocessors.is_shown())
+        self.assertFalse(wyes.Information.ignored_preprocessors.is_shown())
+        self.assertFalse(wfit.Information.ignored_preprocessors.is_shown())
+
+        pp = continuize.Continuize()
+        self.send_signal(wno.Inputs.preprocessor, pp)
+        self.send_signal(wyes.Inputs.preprocessor, pp)
+        self.send_signal(wfit.Inputs.preprocessor, pp)
+
+        self.assertFalse(wno.Information.ignored_preprocessors.is_shown())
+        self.assertTrue(wyes.Information.ignored_preprocessors.is_shown())
+        self.assertFalse(wfit.Information.ignored_preprocessors.is_shown())
+
+        self.send_signal(wno.Inputs.preprocessor, None)
+        self.send_signal(wyes.Inputs.preprocessor, None)
+        self.send_signal(wfit.Inputs.preprocessor, None)
+
+        self.assertFalse(wno.Information.ignored_preprocessors.is_shown())
+        self.assertFalse(wyes.Information.ignored_preprocessors.is_shown())
+        self.assertFalse(wfit.Information.ignored_preprocessors.is_shown())
+
+    def test_multiple_sends(self):
+        class TestLearner(Learner):
+            name = "Test"
+            __returns__ = Mock()
+
+        class TestWidget(OWBaseLearner):
+            name = "Test"
+            LEARNER = TestLearner
+
+        widget = self.create_widget(TestWidget)
+        pp = continuize.Continuize()
+        with patch.object(widget.Outputs.learner, "send") as model_send, \
+                patch.object(widget.Outputs.model, "send") as learner_send:
+            self.send_signals([(widget.Inputs.data, self.iris),
+                               (widget.Inputs.preprocessor, pp)])
+            learner_send.assert_called_once()
+            model_send.assert_called_once()