Merge pull request #4081 from PrimozGodec/fix-predictor

[FIX] Model's data_to_model_domain supports sparse matrix

Author: janezd

Orange/base.py

@@ -2,7 +2,6 @@
 import itertools
 from collections import Iterable
 import re
-import threading
 
 import numpy as np
 import scipy
@@ -13,6 +12,7 @@
 from Orange.data.util import one_hot
 from Orange.misc.wrapper_meta import WrapperMeta
 from Orange.preprocess import Continuize, RemoveNaNColumns, SklImpute, Normalize
+from Orange.statistics.util import all_nan
 from Orange.util import Reprable
 
 __all__ = ["Learner", "Model", "SklLearner", "SklModel",
@@ -320,9 +320,9 @@ def data_to_model_domain():
 
             if self.original_domain.attributes != data.domain.attributes \
                     and data.X.size \
-                    and not np.isnan(data.X).all():
+                    and not all_nan(data.X):
                 new_data = data.transform(self.original_domain)
-                if np.isnan(new_data.X).all():
+                if all_nan(new_data.X):
                     raise DomainTransformationError(
                         "domain transformation produced no defined values")
                 return new_data.transform(self.domain)

Orange/statistics/util.py

@@ -659,6 +659,32 @@ def any_nan(x, axis=None):
     return result
 
 
+def all_nan(x, axis=None):
+    """
+    Check if all of the values in a matrix is nan. Works for sparse matrix too.
+    """
+    if not sp.issparse(x):
+        return np.isnan(x).all(axis=axis)
+
+    if axis is None:
+        # when x.nnz < actual shape there are zero values which are not nan
+        return np.prod(x.shape) == x.nnz and np.isnan(x.data).all()
+
+    if axis == 0:
+        x = x.tocsc()
+    elif axis == 1:
+        x = x.tocsr()
+
+    ax = x.ndim - axis - 1
+    axis_len = x.shape[axis]
+    result = np.zeros(x.shape[ax], dtype=bool)
+    for i in range(x.shape[ax]):
+        vals = x.data[x.indptr[i]:x.indptr[i + 1]]
+        result[i] = axis_len == len(vals) and np.isnan(vals).all()
+
+    return result
+
+
 def FDR(p_values: Iterable, dependent=False, m=None, ordered=False) -> Iterable:
     """ `False Discovery Rate <http://en.wikipedia.org/wiki/False_discovery_rate>`_
         correction on a list of p-values.

Orange/tests/test_statistics.py

@@ -11,7 +11,8 @@
 from Orange.data.util import assure_array_dense
 from Orange.statistics.util import bincount, countnans, contingency, digitize, \
     mean, nanmax, nanmean, nanmedian, nanmin, nansum, nanunique, stats, std, \
-    unique, var, nanstd, nanvar, nanmode, nan_to_num, FDR, isnan, any_nan
+    unique, var, nanstd, nanvar, nanmode, nan_to_num, FDR, isnan, any_nan, \
+    all_nan
 from sklearn.utils import check_random_state
 
 
@@ -730,6 +731,51 @@ def test_axis_1_with_nans(self, array):
         np.testing.assert_equal(result, expected)
 
 
+class TestAllNans(unittest.TestCase):
+    def setUp(self) -> None:
+        # pylint: disable=bad-whitespace
+        self.x_with_nans = np.array([
+            [0.,     1.,     0.,     np.nan, 3.,     5.],
+            [np.nan, np.nan, np.nan, np.nan, np.nan, np.nan],
+            [0.,     0.,     0.,     np.nan, 7.,     6.],
+            [0.,     0.,     0.,     np.nan, 7.,     np.nan],
+        ])
+        self.x_no_nans = np.arange(12).reshape((3, 4))
+        self.x_only_nans = (np.ones(12) * np.nan).reshape((3, 4))
+
+    @dense_sparse
+    def test_axis_none_without_nans(self, array):
+        self.assertFalse(all_nan(array(self.x_no_nans)))
+
+    @dense_sparse
+    def test_axis_none_with_nans(self, array):
+        self.assertTrue(all_nan(array(self.x_only_nans)))
+
+    @dense_sparse
+    def test_axis_0_without_nans(self, array):
+        expected = np.array([0, 0, 0, 0], dtype=bool)
+        result = all_nan(array(self.x_no_nans), axis=0)
+        np.testing.assert_equal(result, expected)
+
+    @dense_sparse
+    def test_axis_0_with_nans(self, array):
+        expected = np.array([0, 0, 0, 1, 0, 0], dtype=bool)
+        result = all_nan(array(self.x_with_nans), axis=0)
+        np.testing.assert_equal(result, expected)
+
+    @dense_sparse
+    def test_axis_1_without_nans(self, array):
+        expected = np.array([0, 0, 0], dtype=bool)
+        result = all_nan(array(self.x_no_nans), axis=1)
+        np.testing.assert_equal(result, expected)
+
+    @dense_sparse
+    def test_axis_1_with_nans(self, array):
+        expected = np.array([0, 1, 0, 0], dtype=bool)
+        result = all_nan(array(self.x_with_nans), axis=1)
+        np.testing.assert_equal(result, expected)
+
+
 class TestNanModeFixedInScipy(unittest.TestCase):
 
     @unittest.expectedFailure