OWDataProjectionWidget: Fix reloading sparse data

Author: VesnaT

Orange/data/util.py

@@ -95,6 +95,18 @@ def hstack(arrays):
         return np.hstack(arrays)
 
 
+def array_equal(a1, a2):
+    """array_equal that supports sparse and dense arrays with missing values"""
+    if a1.shape != a2.shape:
+        return False
+    i1, j1, v1 = sp.find(a1)
+    i2, j2, v2 = sp.find(a2)
+    a1 = v1 if sp.issparse(a1) else a1[i2, j2]
+    a2 = v2 if sp.issparse(a2) else a2[i1, j1]
+    index_equal = set(zip(i1, j1)) == set(zip(i2, j2))
+    return index_equal and np.allclose(a1, a2, equal_nan=True)
+
+
 def assure_array_dense(a):
     if sp.issparse(a):
         a = a.toarray()

Orange/tests/test_util.py

@@ -8,8 +8,9 @@
 from Orange.util import export_globals, flatten, deprecated, try_, deepgetattr, \
     OrangeDeprecationWarning
 from Orange.data import Table
-from Orange.data.util import vstack, hstack
+from Orange.data.util import vstack, hstack, array_equal
 from Orange.statistics.util import stats
+from Orange.tests.test_statistics import dense_sparse
 
 SOMETHING = 0xf00babe
 
@@ -118,3 +119,28 @@ def test_stats_sparse(self):
         data = Table("iris")
         sparse_x = sp.csr_matrix(data.X)
         self.assertTrue(stats(data.X).all() == stats(sparse_x).all())
+
+    @dense_sparse
+    def test_array_equal(self, array):
+        a1 = array([[0., 2.], [3., np.nan]])
+        a2 = array([[0., 2.], [3., np.nan]])
+        self.assertTrue(array_equal(a1, a2))
+
+        a3 = np.array([[0., 2.], [3., np.nan]])
+        self.assertTrue(array_equal(a1, a3))
+        self.assertTrue(array_equal(a3, a1))
+
+    @dense_sparse
+    def test_array_not_equal(self, array):
+        a1 = array([[0., 2.], [3., np.nan]])
+        a2 = array([[0., 2.], [4., np.nan]])
+        self.assertFalse(array_equal(a1, a2))
+
+        a3 = array([[0., 2.], [3., np.nan], [4., 5.]])
+        self.assertFalse(array_equal(a1, a3))
+
+    def test_csc_array_equal(self):
+        a1 = sp.csc_matrix(([1, 4, 5], ([0, 0, 1], [0, 2, 2])), shape=(2, 3))
+        a2 = sp.csc_matrix(([5, 1, 4], ([1, 0, 0], [2, 0, 2])), shape=(2, 3))
+        a2[0, 1] = 0  # explicitly setting to 0
+        self.assertTrue(array_equal(a1, a2))

Orange/widgets/visualize/tests/test_owprojectionwidget.py

@@ -173,3 +173,10 @@ def test_get_coordinates_data(self):
         self.assertEqual(len(self.widget.get_coordinates_data()[0]), 9)
         self.widget.valid_data = np.zeros((10,), dtype=bool)
         self.assertIsNone(self.widget.get_coordinates_data()[0])
+
+    def test_sparse_data_reload(self):
+        table = Table("heart_disease").to_sparse()
+        self.widget.setup_plot = Mock()
+        self.send_signal(self.widget.Inputs.data, table)
+        self.send_signal(self.widget.Inputs.data, table)
+        self.widget.setup_plot.assert_called_once()

Orange/widgets/visualize/utils/widget.py

@@ -8,7 +8,7 @@
 from Orange.data import (
     Table, ContinuousVariable, Domain, Variable, StringVariable
 )
-from Orange.data.util import get_unique_names
+from Orange.data.util import get_unique_names, array_equal
 from Orange.data.sql.table import SqlTable
 from Orange.preprocess.preprocess import Preprocess, ApplyDomain
 from Orange.statistics.util import bincount
@@ -432,9 +432,7 @@ def set_data(self, data):
         self.use_context()
         self._invalidated = not (
             data_existed and self.data is not None and
-            effective_data.X.shape == self.effective_data.X.shape and
-            np.allclose(effective_data.X,
-                        self.effective_data.X, equal_nan=True))
+            array_equal(effective_data.X, self.effective_data.X))
         if self._invalidated:
             self.clear()
         self.enable_controls()