fix interface bug of block_expand_layer and add unittest (PaddlePaddl…

…e#265)

* fix interface bug of block_expand_layer and add unittest

* auto compute num_channels

* default value of num_channels is None

* adjust input order of block_expand

python/paddle/trainer_config_helpers/layers.py

@@ -54,7 +54,7 @@
            'cross_entropy_with_selfnorm', 'cross_entropy',
            'multi_binary_label_cross_entropy',
            'rank_cost', 'lambda_cost', 'huber_cost',
-           # 'block_expand_layer',  # TODO(yuyang18): this layer is not correct
+           'block_expand_layer',
            'maxout_layer', 'out_prod_layer', 'print_layer'
            ]
 
@@ -3284,18 +3284,18 @@ def linear_comb_layer(weights, vectors, size=None, name=None,
 @wrap_name_default()
 @layer_support()
 def block_expand_layer(input,
-                       channel=0,
                        block_x=0,
                        block_y=0,
                        stride_x=0,
                        stride_y=0,
                        padding_x=0,
                        padding_y=0,
+                       num_channels=None,
                        name=None,
                        layer_attr=None):
     """
     Expand feature map to minibatch matrix.
-       - matrix width is: block_y * block_x * channel
+       - matrix width is: block_y * block_x * num_channels
        - matirx height is: outputH * outputW
 
     .. math::
@@ -3307,24 +3307,24 @@ def block_expand_layer(input,
     The expand method is the same with ExpandConvLayer, but saved the transposed
     value. After expanding, output.sequenceStartPositions will store timeline.
     The number of time steps are outputH * outputW and the dimension of each
-    time step is block_y * block_x * channel. This layer can be used after
+    time step is block_y * block_x * num_channels. This layer can be used after
     convolution neural network, and before recurrent neural network.
 
     The simple usage is:
 
     .. code-block:: python
 
        block_expand = block_expand_layer(input,
-                                         channel=128,
+                                         num_channels=128,
                                          stride_x=1,
                                          stride_y=1,
                                          block_x=1,
                                          block_x=3)
 
     :param input: The input layer.
     :type input: LayerOutput
-    :param channel: The channel number of input layer.
-    :type channel: int
+    :param num_channels: The channel number of input layer.
+    :type num_channels: int|None
     :param block_x: The width of sub block.
     :type block_x: int
     :param block_y: The width of sub block.
@@ -3344,16 +3344,18 @@ def block_expand_layer(input,
     :return: LayerOutput object.
     :rtype: LayerOutput
     """
+    if num_channels is None:
+        assert input.num_filters is not None
+        num_channels = input.num_filters
     Layer(name=name,
-          input=Input(input.name,
-                      block_expand=BlockExpand(channels=channel,
-                                               block_x=block_x,
-                                               block_y=block_y,
-                                               stride_x=stride_x,
-                                               stride_y=stride_y,
-                                               padding_x=padding_x,
-                                               padding_y=padding_y)
-                      ),
+          inputs=Input(input.name,
+                       block_expand=BlockExpand(channels=num_channels,
+                                                block_x=block_x,
+                                                block_y=block_y,
+                                                stride_x=stride_x,
+                                                stride_y=stride_y,
+                                                padding_x=padding_x,
+                                                padding_y=padding_y)),
           type=LayerType.BLOCK_EXPAND,
           **ExtraLayerAttribute.to_kwargs(layer_attr)
           )

python/paddle/trainer_config_helpers/tests/configs/check.md5

@@ -12,7 +12,7 @@ a5d9259ff1fd7ca23d0ef090052cb1f2  last_first_seq.protostr
 8bb44e1e5072d0c261572307e7672bda  test_grumemory_layer.protostr
 1f3510672dce7a9ed25317fc58579ac7  test_hsigmoid.protostr
 d350bd91a0dc13e854b1364c3d9339c6  test_lstmemory_layer.protostr
-6fa59551808ee7012bbd24f757e782d2  test_maxout.protostr
+5433ed33d4e7414eaf658f2a55946186  test_maxout.protostr
 251a948ba41c1071afcd3d9cf9c233f7  test_ntm_layers.protostr
 e6ff04e70aea27c7b06d808cc49c9497  test_print_layer.protostr
 2a75dd33b640c49a8821c2da6e574577  test_rnn_group.protostr

python/paddle/trainer_config_helpers/tests/configs/test_maxout.py

@@ -25,6 +25,25 @@
                       stride=2,
                       pool_type=MaxPooling())
 
-fc = fc_layer(input=pool, size=384, bias_attr=False)
+conv2 = img_conv_layer(input=pool,
+                       filter_size = 3,
+                       num_channels=32,
+                       num_filters=128,
+                       padding=1,
+                       act=LinearActivation(),
+                       bias_attr=True)
+
+maxout2 = maxout_layer(input=conv,
+                       num_channels=128,
+                       groups=4)
+
+block = block_expand_layer(input=maxout,
+                           num_channels=32,
+                           stride_x=1,
+                           stride_y=1,
+                           block_x=1,
+                           block_y=6)
+
+fc = fc_layer(input=block, size=384, bias_attr=False)
 
 outputs(fc)