Merge pull request #2138 from pavlin-policar/fitter-sklearn-quick

[FIX] Fitter: Fix used_vals and params not being set

Author: janezd

Orange/modelling/ada_boost.py

@@ -2,12 +2,12 @@
 from Orange.ensembles import (
     SklAdaBoostClassificationLearner, SklAdaBoostRegressionLearner
 )
-from Orange.modelling import Fitter
+from Orange.modelling import SklFitter
 
 __all__ = ['SklAdaBoostLearner']
 
 
-class SklAdaBoostLearner(Fitter):
+class SklAdaBoostLearner(SklFitter):
     __fits__ = {'classification': SklAdaBoostClassificationLearner,
                 'regression': SklAdaBoostRegressionLearner}
 

Orange/modelling/base.py

@@ -1,22 +1,9 @@
-from Orange.base import Learner, Model
+import numpy as np
 
+from Orange.base import Learner, Model, SklLearner
 
-class FitterMeta(type):
-    """Ensure that each subclass of the `Fitter` class overrides the `__fits__`
-    attribute with a valid value."""
-    def __new__(mcs, name, bases, attrs):
-        # Check that a fitter implementation defines a valid `__fits__`
-        if any(cls.__name__ == 'Fitter' for cls in bases):
-            fits = attrs.get('__fits__')
-            assert isinstance(fits, dict), '__fits__ must be dict instance'
-            assert fits.get('classification') and fits.get('regression'), \
-                ('`__fits__` property does not define classification '
-                 'or regression learner. Use a simple learner if you don\'t '
-                 'need the functionality provided by Fitter.')
-        return super().__new__(mcs, name, bases, attrs)
 
-
-class Fitter(Learner, metaclass=FitterMeta):
+class Fitter(Learner):
     """Handle multiple types of target variable with one learner.
 
     Subclasses of this class serve as a sort of dispatcher. When subclassing,
@@ -119,3 +106,14 @@ def params(self):
     def get_params(self, problem_type):
         """Access the specific learner params of a given learner."""
         return self.get_learner(problem_type).params
+
+
+class SklFitter(Fitter):
+    def _fit_model(self, data):
+        model = super()._fit_model(data)
+        model.used_vals = [np.unique(y) for y in data.Y[:, None].T]
+        if data.domain.has_discrete_class:
+            model.params = self.get_params(self.CLASSIFICATION)
+        else:
+            model.params = self.get_params(self.REGRESSION)
+        return model

Orange/modelling/knn.py

@@ -1,10 +1,10 @@
 from Orange.classification import KNNLearner as KNNClassification
-from Orange.modelling import Fitter
+from Orange.modelling import SklFitter
 from Orange.regression import KNNRegressionLearner
 
 __all__ = ['KNNLearner']
 
 
-class KNNLearner(Fitter):
+class KNNLearner(SklFitter):
     __fits__ = {'classification': KNNClassification,
                 'regression': KNNRegressionLearner}

Orange/modelling/linear.py

@@ -1,11 +1,11 @@
 from Orange.classification.sgd import SGDClassificationLearner
-from Orange.modelling import Fitter
+from Orange.modelling import SklFitter
 from Orange.regression import SGDRegressionLearner
 
 __all__ = ['SGDLearner']
 
 
-class SGDLearner(Fitter):
+class SGDLearner(SklFitter):
     name = 'sgd'
 
     __fits__ = {'classification': SGDClassificationLearner,

Orange/modelling/neural_network.py

@@ -1,10 +1,10 @@
 from Orange.classification import NNClassificationLearner
-from Orange.modelling import Fitter
+from Orange.modelling import SklFitter
 from Orange.regression import NNRegressionLearner
 
 __all__ = ['NNLearner']
 
 
-class NNLearner(Fitter):
+class NNLearner(SklFitter):
     __fits__ = {'classification': NNClassificationLearner,
                 'regression': NNRegressionLearner}

Orange/modelling/randomforest.py

@@ -1,12 +1,12 @@
 from Orange.base import RandomForestModel
 from Orange.classification import RandomForestLearner as RFClassification
-from Orange.modelling import Fitter
+from Orange.modelling import SklFitter
 from Orange.regression import RandomForestRegressionLearner as RFRegression
 
 __all__ = ['RandomForestLearner']
 
 
-class RandomForestLearner(Fitter):
+class RandomForestLearner(SklFitter):
     name = 'random forest'
 
     __fits__ = {'classification': RFClassification,

Orange/modelling/svm.py

@@ -3,19 +3,19 @@
     LinearSVMLearner as LinearSVCLearner,
     NuSVMLearner as NuSVCLearner,
 )
-from Orange.modelling import Fitter
+from Orange.modelling import SklFitter
 from Orange.regression import SVRLearner, LinearSVRLearner, NuSVRLearner
 
 __all__ = ['SVMLearner', 'LinearSVMLearner', 'NuSVMLearner']
 
 
-class SVMLearner(Fitter):
+class SVMLearner(SklFitter):
     __fits__ = {'classification': SVCLearner, 'regression': SVRLearner}
 
 
-class LinearSVMLearner(Fitter):
+class LinearSVMLearner(SklFitter):
     __fits__ = {'classification': LinearSVCLearner, 'regression': LinearSVRLearner}
 
 
-class NuSVMLearner(Fitter):
+class NuSVMLearner(SklFitter):
     __fits__ = {'classification': NuSVCLearner, 'regression': NuSVRLearner}

Orange/modelling/tree.py

@@ -1,14 +1,14 @@
 from Orange.classification import SklTreeLearner
 from Orange.classification import TreeLearner as ClassificationTreeLearner
-from Orange.modelling import Fitter
+from Orange.modelling import Fitter, SklFitter
 from Orange.regression import TreeLearner as RegressionTreeLearner
 from Orange.regression.tree import SklTreeRegressionLearner
 from Orange.tree import TreeModel
 
 __all__ = ['SklTreeLearner', 'TreeLearner']
 
 
-class SklTreeLearner(Fitter):
+class SklTreeLearner(SklFitter):
     name = 'tree'
 
     __fits__ = {'classification': SklTreeLearner,

Orange/tests/test_fitter.py

@@ -29,16 +29,6 @@ def setUpClass(cls):
         cls.heart_disease = Table('heart_disease')
         cls.housing = Table('housing')
 
-    def test_throws_if_fits_property_is_invalid(self):
-        """The `__fits__` attribute must be an instance of `LearnerTypes`"""
-        with self.assertRaises(Exception):
-            class DummyFitter(Fitter):
-                name = 'dummy'
-                __fits__ = (DummyClassificationLearner, DummyRegressionLearner)
-
-            fitter = DummyFitter()
-            fitter(self.heart_disease)
-
     def test_dispatches_to_correct_learner(self):
         """Based on the input data, it should dispatch the fitting process to
         the appropriate learner"""
@@ -102,15 +92,6 @@ class DummyFitter(Fitter):
         except TypeError:
             self.fail('Fitter did not properly distribute params to learners')
 
-    def test_error_for_data_type_with_no_learner(self):
-        """If we attempt to define a fitter which only handles one data type
-        it makes more sense to simply use a Learner."""
-        with self.assertRaises(AssertionError):
-            class DummyFitter(Fitter):
-                name = 'dummy'
-                __fits__ = {'classification': None,
-                            'regression': DummyRegressionLearner}
-
     def test_correctly_sets_preprocessors_on_learner(self):
         """Fitters have to be able to pass the `use_default_preprocessors` and
         preprocessors down to individual learners"""