OWDistances: Use only binary features in Jaccard

Author: janezd

Orange/distance/__init__.py

@@ -3,4 +3,5 @@
                        SpearmanR, SpearmanRAbsolute, PearsonR, PearsonRAbsolute,
                        Mahalanobis, MahalanobisDistance, Hamming)
 
-from .base import _preprocess, remove_discrete_features, impute
+from .base import (
+    _preprocess, remove_discrete_features, remove_nonbinary_features, impute)

Orange/distance/base.py

@@ -38,6 +38,15 @@ def remove_discrete_features(data):
     return data.transform(new_domain)
 
 
+def remove_nonbinary_features(data):
+    """Remove non-binary columns from the data."""
+    new_domain = Domain(
+        [a for a in data.domain.attributes
+         if a.is_discrete and len(a.values) == 2],
+        data.domain.class_vars,
+        data.domain.metas)
+    return data.transform(new_domain)
+
 def impute(data):
     """Impute missing values."""
     return SklImpute()(data)

Orange/widgets/unsupervised/owdistances.py

@@ -53,12 +53,14 @@ class Outputs:
 
     class Error(OWWidget.Error):
         no_continuous_features = Msg("No numeric features")
+        no_binary_features = Msg("No binary features")
         dense_metric_sparse_data = Msg("{} requires dense data.")
         distances_memory_error = Msg("Not enough memory")
         distances_value_error = Msg("Problem in calculation:\n{}")
 
     class Warning(OWWidget.Warning):
         ignoring_discrete = Msg("Ignoring categorical features")
+        ignoring_nonbinary = Msg("Ignoring non-binary features")
         imputing_data = Msg("Missing values were imputed")
 
     def __init__(self):
@@ -112,6 +114,7 @@ def _check_sparse():
             if issparse(data.X) and not metric.supports_sparse:
                 self.Error.dense_metric_sparse_data(METRICS[self.metric_idx][0])
                 return False
+            return True
 
         def _fix_discrete():
             nonlocal data
@@ -124,18 +127,34 @@ def _fix_discrete():
                     return False
                 self.Warning.ignoring_discrete()
                 data = distance.remove_discrete_features(data)
+            return True
+
+        def _fix_nonbinary():
+            nonlocal data
+            if metric is distance.Jaccard:
+                nbinary = sum(a.is_discrete and len(a.values) == 2
+                              for a in data.domain.attributes)
+                if not nbinary:
+                    self.Error.no_binary_features()
+                    return False
+                elif nbinary < len(data.domain.attributes):
+                    self.Warning.ignoring_nonbinary()
+                    data = distance.remove_nonbinary_features(data)
+            return True
 
         def _fix_missing():
             nonlocal data
             if not metric.supports_missing and bn.anynan(data.X):
                 self.Warning.imputing_data()
                 data = distance.impute(data)
+            return True
 
         self.clear_messages()
         if data is None:
             return
-        for check in (_check_sparse, _fix_discrete, _fix_missing):
-            if check() is False:
+        for check in (_check_sparse,
+                      _fix_discrete, _fix_missing, _fix_nonbinary):
+            if not check():
                 return
         try:
             if metric.supports_normalization and self.normalized_dist:

Orange/widgets/unsupervised/tests/test_owdistances.py

@@ -4,7 +4,8 @@
 
 import numpy as np
 
-from Orange.data import Table
+from Orange.data import Table, Domain
+from Orange import distance
 from Orange.widgets.unsupervised.owdistances import OWDistances, METRICS
 from Orange.widgets.tests.base import WidgetTest
 
@@ -32,8 +33,9 @@ def test_distance_combo(self):
             else:
                 expected = metric(self.iris)
 
-            np.testing.assert_array_equal(
-                expected, self.get_output(self.widget.Outputs.distances))
+            if metric is not distance.Jaccard:
+                np.testing.assert_array_equal(
+                    expected, self.get_output(self.widget.Outputs.distances))
 
     def test_error_message(self):
         """Check if error message appears and then disappears when
@@ -46,6 +48,36 @@ def test_error_message(self):
         self.send_signal(self.widget.Inputs.data, None)
         self.assertFalse(self.widget.Error.no_continuous_features.is_shown())
 
+    def test_jaccard_messages(self):
+        for self.widget.metric_idx, (name, _) in enumerate(METRICS):
+            if name == "Jaccard":
+                break
+        self.send_signal(self.widget.Inputs.data, self.iris)
+        self.assertTrue(self.widget.Error.no_binary_features.is_shown())
+        self.assertFalse(self.widget.Warning.ignoring_nonbinary.is_shown())
+
+        self.send_signal(self.widget.Inputs.data, None)
+        self.assertFalse(self.widget.Error.no_binary_features.is_shown())
+        self.assertFalse(self.widget.Warning.ignoring_nonbinary.is_shown())
+
+        self.send_signal(self.widget.Inputs.data, self.titanic)
+        self.assertFalse(self.widget.Error.no_binary_features.is_shown())
+        self.assertTrue(self.widget.Warning.ignoring_nonbinary.is_shown())
+
+        self.send_signal(self.widget.Inputs.data, None)
+        self.assertFalse(self.widget.Error.no_binary_features.is_shown())
+        self.assertFalse(self.widget.Warning.ignoring_nonbinary.is_shown())
+
+        self.send_signal(self.widget.Inputs.data, self.titanic)
+        self.assertFalse(self.widget.Error.no_binary_features.is_shown())
+        self.assertTrue(self.widget.Warning.ignoring_nonbinary.is_shown())
+
+        dom = self.titanic.domain
+        dom = Domain(dom.attributes[1:], dom.class_var)
+        self.send_signal(self.widget.Inputs.data, self.titanic.transform(dom))
+        self.assertFalse(self.widget.Error.no_binary_features.is_shown())
+        self.assertFalse(self.widget.Warning.ignoring_nonbinary.is_shown())
+
     def test_too_big_array(self):
         """
         Users sees an error message when calculating too large arrays and Orange