Merge pull request #6508 from markotoplak/clean-sklimpute

Clean up SklImpute

Orange/preprocess/preprocess.py

@@ -158,21 +158,18 @@ def __call__(self, data):
         if isinstance(data, SqlTable):
             return Impute()(data)
         imputer = SimpleImputer(strategy=self.strategy)
-        X = imputer.fit_transform(data.X)
+        imputer.fit(data.X)
         # Create new variables with appropriate `compute_value`, but
         # drop the ones which do not have valid `imputer.statistics_`
         # (i.e. all NaN columns). `sklearn.preprocessing.Imputer` already
         # drops them from the transformed X.
-        features = [impute.Average()(data, var, value)
+        features = [var.copy(compute_value=impute.ReplaceUnknowns(var, value))
                     for var, value in zip(data.domain.attributes,
                                           imputer.statistics_)
                     if not np.isnan(value)]
-        assert X.shape[1] == len(features)
         domain = Orange.data.Domain(features, data.domain.class_vars,
                                     data.domain.metas)
         new_data = data.transform(domain)
-        with new_data.unlocked(new_data.X):
-            new_data.X = X
         return new_data
 
 

Orange/tests/test_impute.py

@@ -7,7 +7,7 @@
 import scipy.sparse as sp
 
 from Orange import preprocess
-from Orange.preprocess import impute
+from Orange.preprocess import impute, SklImpute
 from Orange import data
 from Orange.data import Unknown, Table
 
@@ -329,3 +329,44 @@ def test_imputer(self):
         auto = data.Table(test_filename('datasets/imports-85.tab'))
         auto2 = preprocess.Impute()(auto)
         self.assertFalse(np.isnan(auto2.X).any())
+
+
+class TestSklImpute(unittest.TestCase):
+
+    def setUp(self):
+        nan = np.nan
+        X = [
+            [1.0, nan, 0.0],
+            [2.0, 1.0, 3.0],
+            [nan, nan, nan]
+        ]
+        self.imputed_mean = [
+            [1.0, 1.0, 0.0],
+            [2.0, 1.0, 3.0],
+            [1.5, 1.0, 1.5]
+        ]
+        domain = data.Domain((data.ContinuousVariable(n) for n in "ABC"))
+        self.table = data.Table.from_numpy(domain, np.array(X))
+
+    def test_values(self):
+        imputed = SklImpute()(self.table)
+        np.testing.assert_equal(imputed.X, self.imputed_mean)
+
+    def test_sparse(self):
+        sparse = self.table.to_sparse()
+        self.assertTrue(sp.issparse(sparse.X))
+        imputed = SklImpute()(sparse)
+        self.assertTrue(sp.issparse(imputed.X))
+        np.testing.assert_equal(imputed.X.todense(), self.imputed_mean)
+
+    def test_transform(self):
+        imputed = SklImpute()(self.table)
+        transformed = self.table.transform(imputed.domain)
+        np.testing.assert_equal(transformed.X, self.imputed_mean)
+
+    def test_transform_sparse(self):
+        sparse = self.table.to_sparse()
+        imputed = SklImpute()(sparse)
+        self.assertTrue(sp.issparse(sparse.X))
+        transformed = sparse.transform(imputed.domain)
+        np.testing.assert_equal(transformed.X.todense(), self.imputed_mean)

benchmark/bench_preprocess.py

@@ -3,12 +3,12 @@
 import numpy as np
 
 from Orange.data import Domain, Table, ContinuousVariable
-from Orange.preprocess import Normalize
+from Orange.preprocess import Normalize, SklImpute
 
 from .base import Benchmark, benchmark
 
 
-class BenchNormalize(Benchmark):
+class SetUpData:
 
     def setUp(self):
         cols = 1000
@@ -21,6 +21,9 @@ def setUp(self):
             np.random.RandomState(0).randint(0, 2, (rows, len(self.domain.variables))))
         self.normalized_domain = Normalize()(self.table).domain
 
+
+class BenchNormalize(SetUpData, Benchmark):
+
     @benchmark(number=5)
     def bench_normalize_only_transform(self):
         self.table.transform(self.normalized_domain)
@@ -30,3 +33,10 @@ def bench_normalize_only_parameters(self):
         # avoid benchmarking transformation
         with patch("Orange.data.Table.transform", MagicMock()):
             Normalize()(self.table)
+
+
+class BenchSklImpute(SetUpData, Benchmark):
+
+    @benchmark(number=5)
+    def bench_sklimpute(self):
+        SklImpute()(self.table)