Merge pull request #275 from janezd/fix-text-to-columns-dtype

Text to Columns: Fix dtype for empty arrays

orangecontrib/prototypes/widgets/owtexttocolumns.py

@@ -31,7 +31,8 @@ def __call__(self, data):
         column = data.get_column(self.attr)
         values = [{ss.strip() for ss in s.split(self.delimiter)}
                   for s in column]
-        return {v: np.array([i for i, xs in enumerate(values) if v in xs])
+        return {v: np.array([i for i, xs in enumerate(values) if v in xs],
+                            dtype=int)
                 for v in self.new_values}
 
     def __eq__(self, other):

orangecontrib/prototypes/widgets/tests/test_owtexttocolumns.py

@@ -47,6 +47,17 @@ def test_split_column(self):
         np.testing.assert_equal(shared["f o"], [0, 1])
         np.testing.assert_equal(shared["o"], [2])
 
+    def test_no_known_values(self):
+        sc = SplitColumn(self.data, self.data.domain.metas[0], ",")
+        data = Table.from_numpy(
+            self.data.domain, np.zeros((3, 1)), None,
+            np.array([["x"] * 2] * 3))
+        shared = sc(data)
+        for attr in ("a", "bbb", "d"):
+            self.assertEqual(shared[attr].size, 0)
+            oh = OneHotStrings(sc, attr)
+            np.testing.assert_equal(oh(data), [0, 0, 0])
+
     def test_one_hot_strings(self):
         attr = self.data.domain.metas[0]
         sc = SplitColumn(self.data, attr, ",")