add gated linear unit into config helper.

doc/api/v2/config/layer.rst

@@ -474,6 +474,11 @@ prelu
 ..  autoclass:: paddle.v2.layer.prelu
     :noindex:
 
+gated_unit
+-----------
+..  autoclass:: paddle.v2.layer.gated_unit
+    :noindex:
+
 Detection output Layer
 ======================
 

python/paddle/trainer_config_helpers/layers.py

@@ -126,6 +126,7 @@ __all__ = [
     'row_conv_layer',
     'dropout_layer',
     'prelu_layer',
+    'gated_unit_layer',
 ]
 
 
@@ -5862,7 +5863,7 @@ def prelu_layer(input,
     :rtype: LayerOutput
     """
 
-    assert isinstance(input, LayerOutput), 'prelu_layer only accepts one input'
+    assert isinstance(input, LayerOutput), 'prelu_layer accepts only one input.'
     assert isinstance(param_attr, ParameterAttribute)
 
     l = Layer(
@@ -5876,3 +5877,96 @@ def prelu_layer(input,
         layer_type=LayerType.PRELU,
         parents=input,
         size=l.config.size)
+
+
+@layer_support(ERROR_CLIPPING, DROPOUT)
+@wrap_name_default()
+@wrap_act_default(act=LinearActivation())
+def gated_unit_layer(input,
+                     size,
+                     act=None,
+                     name=None,
+                     gate_attr=None,
+                     gate_bias_attr=True,
+                     gate_param_attr=None,
+                     inproj_param_attr=None,
+                     inproj_bias_attr=True,
+                     inproj_layer_attr=None,
+                     layer_attr=None):
+    """
+    The gated unit layer implements a simple gating mechanism over the input.
+    The input :math:`X` is first projected into a new space :math:`X'`, and
+    it is also used to produce a gate weight :math:`\sigma`. Element-wise
+    prodict between :match:`X'` and :math:`\sigma` is finally returned.
+
+    Reference:
+        Language Modeling with Gated Convolutional Networks
+        https://arxiv.org/abs/1612.08083
+
+    .. math::
+       y=\\text{act}(X \cdot W + b)\otimes \sigma(X \cdot V + c)
+
+    The example usage is:
+
+    .. code-block:: python
+        gated_unit = gated_unit_layer(size=128, input=input_layer))
+
+    :param input: input for this layer.
+    :type input: LayerOutput
+    :param size: output size of the gated unit.
+    :type size: int
+    :param act: activation type of the projected input.
+    :type act: BaseActivation
+    :param name: name of this layer.
+    :type name: basestring
+    :param gate_attr: Attributes to tune the gate output, for example, error
+        clipping threshold, dropout and so on. See ExtraLayerAttribute for
+        more details.
+    :type gate_attr: ExtraLayerAttribute|None
+    :param gate_bias_attr: Attributes to tune the learnable bias of the gate.
+    :type gate_bias_attr: ParameterAttribute|None
+    :param gate_param_attr: Attributes to tune the learnable projected matrix
+        parameter of the gate.
+    :type gate_param_attr: ParameterAttribute|None
+    :param inproj_param_attr: Attributes to tune the learnable parameter of
+        the projection of input.
+    :type inproj_param_attr: ParameterAttribute|None
+    :param inproj_layer_attr: Attributes to the tune the projected input, for
+        example, error clipping threshold, dropout and so on. See
+        ExtraLayerAttribute for more details.
+    :type inproj_layer_attr: ExtraLayerAttribute|None
+    :param inproj_bias_attr: Attributes to tune the learnable bias of
+        projection of the input.
+    :type inproj_bias_attr: ParameterAttribute|None
+    :param layer_attr: Attributes to tune the final output of the gated unit,
+        for example, error clipping threshold, dropout and so on. See
+        ExtraLayerAttribute for more details.
+    :type layer_attr: ExtraLayerAttribute|None
+    :return: LayerOutput object.
+    :rtype: LayerOutput
+    """
+
+    assert isinstance(
+        input, LayerOutput), 'The gated linear unit accepts only one input.'
+
+    input_proj = fc_layer(
+        input=input,
+        name="%s_input_proj" % name,
+        size=size,
+        act=act,
+        param_attr=inproj_param_attr,
+        layer_attr=inproj_layer_attr,
+        bias_attr=inproj_bias_attr)
+
+    gate = fc_layer(
+        size=size,
+        name="%s_gate" % name,
+        act=SigmoidActivation(),
+        input=input,
+        param_attr=gate_param_attr,
+        layer_attr=gate_attr,
+        bias_attr=gate_bias_attr)
+    return mixed_layer(
+        name="%s_gated_act" % name,
+        input=dotmul_operator(input_proj, gate),
+        layer_attr=layer_attr)

python/paddle/trainer_config_helpers/tests/configs/file_list.sh

@@ -7,6 +7,6 @@ test_rnn_group shared_fc shared_lstm shared_gru test_cost_layers_with_weight
 test_spp_layer test_bilinear_interp test_maxout test_bi_grumemory math_ops
 test_seq_concat_reshape test_pad test_smooth_l1 test_multiplex_layer
 test_prelu_layer test_row_conv test_detection_output_layer test_multibox_loss_layer
-test_recursive_topology)
+test_recursive_topology test_gated_unit_layer)
 
 export whole_configs=(test_split_datasource)

python/paddle/trainer_config_helpers/tests/configs/protostr/test_gated_unit_layer.protostr

@@ -0,0 +1,106 @@
+type: "nn"
+layers {
+  name: "input"
+  type: "data"
+  size: 256
+  active_type: ""
+}
+layers {
+  name: "__gated_unit_layer_0___input_proj"
+  type: "fc"
+  size: 512
+  active_type: "tanh"
+  inputs {
+    input_layer_name: "input"
+    input_parameter_name: "___gated_unit_layer_0___input_proj.w0"
+  }
+  bias_parameter_name: "___gated_unit_layer_0___input_proj.wbias"
+  error_clipping_threshold: 100.0
+}
+layers {
+  name: "__gated_unit_layer_0___gate"
+  type: "fc"
+  size: 512
+  active_type: "sigmoid"
+  inputs {
+    input_layer_name: "input"
+    input_parameter_name: "___gated_unit_layer_0___gate.w0"
+  }
+  bias_parameter_name: "___gated_unit_layer_0___gate.wbias"
+  error_clipping_threshold: 100.0
+}
+layers {
+  name: "__gated_unit_layer_0___gated_act"
+  type: "mixed"
+  size: 512
+  active_type: ""
+  inputs {
+    input_layer_name: "__gated_unit_layer_0___input_proj"
+  }
+  inputs {
+    input_layer_name: "__gated_unit_layer_0___gate"
+  }
+  error_clipping_threshold: 100.0
+  operator_confs {
+    type: "dot_mul"
+    input_indices: 0
+    input_indices: 1
+    input_sizes: 512
+    input_sizes: 512
+    output_size: 512
+    dotmul_scale: 1
+  }
+}
+parameters {
+  name: "___gated_unit_layer_0___input_proj.w0"
+  size: 131072
+  initial_mean: 0.0
+  initial_std: 0.0001
+  dims: 256
+  dims: 512
+  initial_strategy: 0
+  initial_smart: false
+}
+parameters {
+  name: "___gated_unit_layer_0___input_proj.wbias"
+  size: 512
+  initial_mean: 0.0
+  initial_std: 1
+  dims: 1
+  dims: 512
+  initial_strategy: 0
+  initial_smart: false
+}
+parameters {
+  name: "___gated_unit_layer_0___gate.w0"
+  size: 131072
+  initial_mean: 0.0
+  initial_std: 0.0001
+  dims: 256
+  dims: 512
+  initial_strategy: 0
+  initial_smart: false
+}
+parameters {
+  name: "___gated_unit_layer_0___gate.wbias"
+  size: 512
+  initial_mean: 0.0
+  initial_std: 1
+  dims: 1
+  dims: 512
+  initial_strategy: 0
+  initial_smart: false
+}
+input_layer_names: "input"
+output_layer_names: "__gated_unit_layer_0___gated_act"
+sub_models {
+  name: "root"
+  layer_names: "input"
+  layer_names: "__gated_unit_layer_0___input_proj"
+  layer_names: "__gated_unit_layer_0___gate"
+  layer_names: "__gated_unit_layer_0___gated_act"
+  input_layer_names: "input"
+  output_layer_names: "__gated_unit_layer_0___gated_act"
+  is_recurrent_layer_group: false
+}
+

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

@@ -0,0 +1,16 @@
+from paddle.trainer_config_helpers import *
+
+data = data_layer(name='input', size=256)
+glu = gated_unit_layer(
+    size=512,
+    input=data,
+    act=TanhActivation(),
+    gate_param_attr=ParamAttr(initial_std=1e-4),
+    gate_attr=ExtraLayerAttribute(error_clipping_threshold=100.0),
+    gate_bias_attr=ParamAttr(initial_std=1),
+    inproj_param_attr=ParamAttr(initial_std=1e-4),
+    inproj_layer_attr=ExtraLayerAttribute(error_clipping_threshold=100.0),
+    inproj_bias_attr=ParamAttr(initial_std=1),
+    layer_attr=ExtraLayerAttribute(error_clipping_threshold=100.0))
+
+outputs(glu)