Fix impute.Model for derived domains

The compute_value was missing transformation into the variable space it
was working upon.

Orange/preprocess/impute.py

@@ -172,7 +172,7 @@ def copy(self):
         return FixedValueByType(*self.defaults.values())
 
 
-class ReplaceUnknownsModel(Reprable):
+class ReplaceUnknownsModel(Transformation):
     """
     Replace unknown values with predicted values using a `Orange.base.Model`
 
@@ -185,15 +185,14 @@ class ReplaceUnknownsModel(Reprable):
     """
     def __init__(self, variable, model):
         assert model.domain.class_var == variable
-        self.variable = variable
+        super().__init__(variable)
         self.model = model
 
     def __call__(self, data):
         if isinstance(data, Orange.data.Instance):
             data = Orange.data.Table.from_list(data.domain, [data])
         domain = data.domain
-        column = data.get_column(self.variable, copy=True)
-
+        column = data.transform(self._target_domain).get_column(self.variable, copy=True)
         mask = np.isnan(column)
         if not np.any(mask):
             return column

Orange/tests/test_impute.py

@@ -9,7 +9,7 @@
 from Orange import preprocess
 from Orange.preprocess import impute, SklImpute
 from Orange import data
-from Orange.data import Unknown, Table
+from Orange.data import Unknown, Table, Domain
 
 from Orange.classification import MajorityLearner, SimpleTreeLearner
 from Orange.regression import MeanLearner
@@ -293,6 +293,27 @@ def test_bad_domain(self):
         self.assertRaises(ValueError, imputer, data=table,
                           variable=table.domain[0])
 
+    def test_missing_imputed_columns(self):
+        housing = Table("housing")
+
+        learner = SimpleTreeLearner(min_instances=10, max_depth=10)
+        method = preprocess.impute.Model(learner)
+
+        ivar = method(housing, housing.domain.attributes[0])
+        imputed = housing.transform(
+            Domain([ivar],
+                   housing.domain.class_var)
+        )
+        removed_imputed = imputed.transform(
+            Domain([], housing.domain.class_var))
+
+        r = removed_imputed.transform(imputed.domain)
+
+        no_class = removed_imputed.transform(Domain(removed_imputed.domain.attributes, None))
+        impute_model_prediction_for_unknowns = ivar.compute_value.model(no_class[0])
+
+        np.testing.assert_equal(r.X, impute_model_prediction_for_unknowns)
+
 
 class TestRandom(unittest.TestCase):
     def test_replacement(self):