Tests for arbitrary inputs should now raise errors.

theano/tensor/nnet/tests/test_abstract_conv.py

@@ -472,13 +472,10 @@ class BaseTestConv2d(BaseTestConv):
         filter_shape = (2, 1, 3, 3)
         for output_shape in [(2, 2, 8, 8), (2, 2, 9, 9), (2, 2, 12, 12)]:
             for border_mode in ["valid", "half", "full"]:
-                # is this output shape large enough?
-                min_output_shape = self.get_output_shape(
-                    input_shape, filter_shape, self.default_subsamples,
-                    border_mode, self.default_filters_dilations)
-                if not all(o >= min_o for (o, min_o) in zip(output_shape, min_output_shape)):
-                    continue
-                for provide_shape in self.provide_shape:
+                computed_shape = get_conv_output_shape(
+                    input_shape, filter_shape, border_mode, self.default_subsamples, self.default_filters_dilations)
+                # is this a valid combination?
+                if tuple(computed_shape) == output_shape:
                     yield (self.tcase_gi,
                            input_shape,
                            filter_shape,
@@ -486,8 +483,21 @@ class BaseTestConv2d(BaseTestConv):
                            self.default_subsamples,
                            border_mode,
                            True,
-                           provide_shape,
-                           self.default_filters_dilations)
+                           True,
+                           self.default_filters_dilations,
+                           False)
+                else:
+                    # expect an error
+                    yield (self.tcase_gi,
+                           input_shape,
+                           filter_shape,
+                           output_shape,
+                           self.default_subsamples,
+                           border_mode,
+                           True,
+                           True,
+                           self.default_filters_dilations,
+                           True)
 
     def test_gradinput_impossible_output_shapes(self):
         for i in range(1, 20):
@@ -505,6 +515,7 @@ class BaseTestConv2d(BaseTestConv):
                             # outputs that are too large or too small should be rejected
                             for o in (-3, -2, -1, 1, 2, 3):
                                 output_shape = (1, 1, computed_shape[2] + o, computed_shape[3] + o)
+                                # expect an error
                                 yield (self.tcase_gi,
                                        image_shape,
                                        kernel_shape,
@@ -822,13 +833,11 @@ class BaseTestConv3d(BaseTestConv):
         filter_shape = (1, 1, 3, 3, 3)
         for output_shape in [(2, 1, 8, 8, 8), (2, 1, 9, 9, 9), (2, 1, 12, 12, 12)]:
             for border_mode in ["valid", "half", "full"]:
-                # is this output shape large enough?
-                min_output_shape = self.get_output_shape(
-                    input_shape, filter_shape, self.default_subsamples,
-                    border_mode, self.default_filters_dilations)
-                if not all(o >= min_o for (o, min_o) in zip(output_shape, min_output_shape)):
-                    continue
-                for provide_shape in self.provide_shape:
+                # compute the output that these inputs and parameters would produce
+                computed_shape = get_conv_output_shape(
+                    input_shape, filter_shape, border_mode, self.default_subsamples, self.default_filters_dilations)
+                # is this a valid combination?
+                if tuple(computed_shape) == output_shape:
                     yield (self.tcase_gi,
                            input_shape,
                            filter_shape,
@@ -836,8 +845,21 @@ class BaseTestConv3d(BaseTestConv):
                            self.default_subsamples,
                            border_mode,
                            True,
-                           provide_shape,
-                           self.default_filters_dilations)
+                           True,
+                           self.default_filters_dilations,
+                           False)
+                else:
+                    # expect an error
+                    yield (self.tcase_gi,
+                           input_shape,
+                           filter_shape,
+                           output_shape,
+                           self.default_subsamples,
+                           border_mode,
+                           True,
+                           True,
+                           self.default_filters_dilations,
+                           True)
 
     def test_gradinput_impossible_output_shapes(self):
         for i in range(1, 20):