Adding tests for coverage of missing MLP functionality, AE officially…

… disabled.

.travis.yml

@@ -18,7 +18,7 @@ install:
   - pip install -q argparse coveralls lasagne
 
 script:
-  - echo -e "[run]\nomit=\n    sknn/backend/pylearn2/pywrap2.py\n    sknn/backend/deepy/*\n    sknn/backend/__init__.py*\n" > .coveragerc
+  - echo -e "[run]\nomit=\n    *ae.py*\n    sknn/backend/deepy/*\n    sknn/backend/__init__.py*\n" > .coveragerc
   - nosetests --with-coverage --cover-package=sknn -v sknn.tests
 
 after_success:

sknn/backend/lasagne/mlp.py

@@ -194,6 +194,11 @@ def _initialize_impl(self, X, y=None):
                                 test_size=self.valid_size,
                                 random_state=self.random_state)
             self.valid_set = X_v, y_v
+
+        if self.valid_set and self.is_convolution:
+            X_v, y_v = self.valid_set
+            if X_v.shape[-2:] != X.shape[-2:]:
+                self.valid_set = numpy.transpose(X_v, (0, 3, 1, 2)), y_v
 
         params = []
         for spec, mlp_layer in zip(self.layers, self.mlp):
@@ -204,9 +209,6 @@ def _initialize_impl(self, X, y=None):
         return X, y
 
     def _predict_impl(self, X):
-        if not self.is_initialized:
-            self._initialize_impl(X)
-
         if self.is_convolution:
             X = numpy.transpose(X, (0, 3, 1, 2))
         return self.f(X)

sknn/tests/test_classifier.py

@@ -14,6 +14,9 @@ class TestClassifierFunctionality(unittest.TestCase):
     def setUp(self):
         self.nn = MLPC(layers=[L("Softmax")], n_iter=1)
 
+    def test_IsClassifier(self):
+        assert_true(self.nn.is_classifier)
+
     def test_FitAutoInitialize(self):
         a_in, a_out = numpy.zeros((8,16)), numpy.random.randint(0, 5, (8,))
         self.nn.fit(a_in, a_out)

sknn/tests/test_conv.py

@@ -65,14 +65,25 @@ def test_HorizontalKernel(self):
                 L("Linear")],
             n_iter=1))
 
-    def test_ValidationSet(self):
+    def test_ValidationSize(self):
         self._run(MLPR(
             layers=[
                 C("Tanh", channels=4, kernel_shape=(3,3)),
                 L("Linear")],
             n_iter=1,
             valid_size=0.5))
 
+    def test_ValidationSet(self):
+        v_in = numpy.zeros((8,32,16,1))
+        v_out = numpy.zeros((8,4))
+
+        self._run(MLPR(
+            layers=[
+                C("Tanh", channels=4, kernel_shape=(3,3)),
+                L("Linear")],
+            n_iter=1,
+            valid_set=(v_in, v_out)))
+
     def test_MultipleLayers(self):
         self._run(MLPR(
             layers=[

sknn/tests/test_linear.py

@@ -17,6 +17,9 @@ def setUp(self):
 
     def test_LifeCycle(self):
         del self.nn
+
+    def test_IsNotClassifier(self):
+        assert_false(self.nn.is_classifier)
 
     def test_PredictNoOutputUnitsAssertion(self):
         a_in = numpy.zeros((8,16))

sknn/tests/test_output.py

@@ -26,7 +26,7 @@ def test_MeanSquaredErrorLinear(self):
         self._run(nn)
 
     @unittest.skipIf(sknn.backend.name != 'lasagne', 'only lasagne')
-    def test_MeanSquaredErrorLinear(self):
+    def test_CategoricalCrossEntropyLinear(self):
         nn = MLPR(layers=[L("Softmax")], loss_type='mcc', n_iter=1)
         self._run(nn)