Merge pull request #3458 from VesnaT/fix_rank

[FIX] Rank: Fix crash on dataset with missing values

Author: lanzagar

Orange/preprocess/fss.py

@@ -8,7 +8,6 @@
 import Orange
 from Orange.util import Reprable
 from Orange.preprocess.preprocess import Preprocess
-from Orange.preprocess.score import ANOVA, GainRatio, UnivariateLinearRegression
 
 __all__ = ["SelectBestFeatures", "RemoveNaNColumns", "SelectRandomFeatures"]
 
@@ -57,6 +56,10 @@ def __call__(self, data):
             discr_ratio = (sum(a.is_discrete
                                for a in data.domain.attributes)
                            / len(data.domain.attributes))
+
+            from Orange.preprocess.score import ANOVA, GainRatio, \
+                UnivariateLinearRegression
+
             if data.domain.has_discrete_class:
                 if discr_ratio >= 0.5:
                     method = GainRatio()

Orange/preprocess/score.py

@@ -8,6 +8,7 @@
 from Orange.data import Domain, Variable, DiscreteVariable, ContinuousVariable
 from Orange.data.filter import HasClass
 from Orange.misc.wrapper_meta import WrapperMeta
+from Orange.preprocess.fss import RemoveNaNColumns
 from Orange.preprocess.preprocess import Discretize, SklImpute
 from Orange.preprocess.util import _RefuseDataInConstructor
 from Orange.statistics import contingency, distribution
@@ -66,6 +67,7 @@ def __call__(self, data, feature=None):
             f = data.domain[feature]
             data = data.transform(Domain([f], data.domain.class_vars))
 
+        orig_domain = data.domain
         for pp in self.preprocessors:
             data = pp(data)
 
@@ -76,7 +78,14 @@ def __call__(self, data, feature=None):
                     .format(self.friendly_name,
                             self._friendly_vartype_name(type(var))))
 
-        return self.score_data(data, feature)
+        if feature is not None:
+            return self.score_data(data, feature)
+
+        scores = np.full(len(orig_domain.attributes), np.nan)
+        names = [a.name for a in data.domain.attributes]
+        mask = np.array([a.name in names for a in orig_domain.attributes])
+        scores[mask] = self.score_data(data, feature)
+        return scores
 
     def score_data(self, data, feature):
         raise NotImplementedError
@@ -340,6 +349,7 @@ class ReliefF(Scorer):
     class_type = DiscreteVariable
     supports_sparse_data = False
     friendly_name = "ReliefF"
+    preprocessors = Scorer.preprocessors + [RemoveNaNColumns()]
 
     def __init__(self, n_iterations=50, k_nearest=10, random_state=None):
         self.n_iterations = n_iterations
@@ -374,6 +384,7 @@ class RReliefF(Scorer):
     class_type = ContinuousVariable
     supports_sparse_data = False
     friendly_name = "RReliefF"
+    preprocessors = Scorer.preprocessors + [RemoveNaNColumns()]
 
     def __init__(self, n_iterations=50, k_nearest=50, random_state=None):
         self.n_iterations = n_iterations

Orange/widgets/data/tests/test_owrank.py

@@ -10,7 +10,7 @@
 from Orange.regression import LinearRegressionLearner
 from Orange.projection import PCA
 from Orange.widgets.data.owrank import OWRank, ProblemType, CLS_SCORES, REG_SCORES
-from Orange.widgets.tests.base import WidgetTest
+from Orange.widgets.tests.base import WidgetTest, datasets
 from Orange.widgets.widget import AttributeList
 
 
@@ -347,3 +347,9 @@ def test_no_attributes(self):
         self.assertTrue(self.widget.Error.no_attributes.is_shown())
         self.send_signal(self.widget.Inputs.data, data)
         self.assertFalse(self.widget.Error.no_attributes.is_shown())
+
+    def test_dataset(self):
+        for method in CLS_SCORES + REG_SCORES:
+            self._get_checkbox(method.shortname).setChecked(True)
+        for ds in datasets.datasets():
+            self.send_signal(self.widget.Inputs.data, ds)