Esempio n. 1
0
    def test_transpose_undefined_input_shape(self):
        network = layers.Transpose((1, 0, 2))
        self.assertShapesEqual(network.input_shape, None)
        self.assertShapesEqual(network.output_shape, (None, None, None))

        network = layers.Transpose((1, 0))
        self.assertShapesEqual(network.input_shape, None)
        self.assertShapesEqual(network.output_shape, (None, None))
Esempio n. 2
0
    def test_transpose_repr(self):
        layer = layers.Transpose((0, 2, 1))
        self.assertEqual(
            "Transpose((0, 2, 1), name='transpose-1')",
            str(layer))

        layer = layers.Transpose((0, 2, 1), name='test')
        self.assertEqual(
            "Transpose((0, 2, 1), name='test')",
            str(layer))
Esempio n. 3
0
    def test_transpose_exceptions(self):
        with self.assertRaisesRegexp(ValueError, "cannot be used"):
            layers.join(
                layers.Input((7, 11)),
                layers.Transpose([2, 0]),  # cannot use 0 index (batch dim)
            )

        with self.assertRaisesRegexp(LayerConnectionError, "at least 3"):
            layers.join(
                layers.Input(20),
                layers.Transpose([2, 1]),
            )
Esempio n. 4
0
 def test_transpose_exceptions(self):
     error_message = "Cannot apply transpose operation to the input"
     with self.assertRaisesRegexp(LayerConnectionError, error_message):
         layers.join(
             layers.Input(20),
             layers.Transpose((0, 2, 1)),
         )
Esempio n. 5
0
    def test_transpose_unknown_input_dim(self):
        network = layers.join(
            layers.Input((None, 10, 20)),
            layers.Transpose((0, 2, 1, 3)),
        )
        self.assertShapesEqual(network.output_shape, (None, 10, None, 20))

        value = asfloat(np.random.random((12, 100, 10, 20)))
        output_value = self.eval(network.output(value))
        self.assertEqual(output_value.shape, (12, 10, 100, 20))

        value = asfloat(np.random.random((12, 33, 10, 20)))
        output_value = self.eval(network.output(value))
        self.assertEqual(output_value.shape, (12, 10, 33, 20))
Esempio n. 6
0
    def test_transpose_unknown_input_dim(self):
        conn = layers.join(
            layers.Input((None, 10, 20)),
            layers.Transpose([2, 1, 3]),
        )
        self.assertEqual(conn.output_shape, (10, None, 20))

        value = asfloat(np.random.random((12, 100, 10, 20)))
        output_value = self.eval(conn.output(value))
        self.assertEqual(output_value.shape, (12, 10, 100, 20))

        value = asfloat(np.random.random((12, 33, 10, 20)))
        output_value = self.eval(conn.output(value))
        self.assertEqual(output_value.shape, (12, 10, 33, 20))
Esempio n. 7
0
 def test_simple_transpose(self):
     network = layers.join(
         layers.Input((7, 11)),
         layers.Transpose((0, 2, 1)),
     )
     self.assertShapesEqual(network.output_shape, (None, 11, 7))
Esempio n. 8
0
 def test_simple_transpose(self):
     conn = layers.join(
         layers.Input((7, 11)),
         layers.Transpose([2, 1]),
     )
     self.assertEqual(conn.output_shape, (11, 7))