Merge pull request #598 from markotoplak/predictmiss

Ability to learn without and predict wih missing values

Orange/base.py

@@ -214,7 +214,7 @@ class SklLearner(Learner, metaclass=WrapperMeta):
     name = 'skl learner'
     preprocessors = [Continuize(),
                      RemoveNaNColumns(),
-                     SklImpute(force=False)]
+                     SklImpute()]
 
     @property
     def params(self):

Orange/preprocess/preprocess.py

@@ -148,17 +148,13 @@ def __call__(self, data):
 class SklImpute(Preprocess):
     __wraps__ = skl_preprocessing.Imputer
 
-    def __init__(self, strategy='mean', force=True):
+    def __init__(self, strategy='mean'):
         self.strategy = strategy
-        self.force = force
 
     def __call__(self, data):
         from Orange.data.sql.table import SqlTable
         if isinstance(data, SqlTable):
             return Impute()(data)
-
-        if not self.force and not np.isnan(data.X).any():
-            return data
         self.imputer = skl_preprocessing.Imputer(strategy=self.strategy)
         X = self.imputer.fit_transform(data.X)
         # Create new variables with appropriate `compute_value`, but

Orange/projection/base.py

@@ -57,7 +57,7 @@ class SklProjector(Projector, metaclass=WrapperMeta):
     _params = {}
     name = 'skl projection'
     preprocessors = [Orange.preprocess.Continuize(),
-                     Orange.preprocess.SklImpute(force=False)]
+                     Orange.preprocess.SklImpute()]
 
     @property
     def params(self):

Orange/regression/linear.py

@@ -106,7 +106,7 @@ class PolynomialLearner(Learner):
     name = 'poly learner'
     preprocessors = [Continuize(),
                      RemoveNaNColumns(),
-                     SklImpute(force=False)]
+                     SklImpute()]
 
     def __init__(self, learner, degree=1, preprocessors=None):
         super().__init__(preprocessors=preprocessors)

Orange/tests/test_classification.py

@@ -141,7 +141,8 @@ class ExpandProbabilitiesTest(unittest.TestCase):
     def prepareTable(self, rows, attr, vars, class_var_domain):
         attributes = ["Feature %i" % i for i in range(attr)]
         classes = ["Class %i" % i for i in range(vars)]
-        attr_vars = [DiscreteVariable(name=a) for a in attributes]
+        attr_vars = [DiscreteVariable(name=a, values=range(2))
+                     for a in attributes]
         class_vars = [DiscreteVariable(name=c,
                                             values=range(class_var_domain))
                       for c in classes]
@@ -219,6 +220,13 @@ def test_continuous(lf):
         assert(all(tree(strlist) == tree(Orange.data.Table(table.domain, strlist))))
 
 
+class UnknownValuesInPrediction(unittest.TestCase):
+    def test_unknown(self):
+        table = Table("iris")
+        tree = LogisticRegressionLearner()(table)
+        tree([1, 2, None])
+
+
 class LearnerAccessibility(unittest.TestCase):
     def setUp(self):
         Variable._clear_all_caches()