"add sequence conv layer"

python/paddle/v2/framework/layers.py

@@ -5,7 +5,7 @@
 
 __all__ = [
     'fc', 'data', 'cross_entropy', 'conv2d', 'pool2d', 'embedding', 'concat',
-    'StaticRNN'
+    'StaticRNN', 'cast'
 ]
 
 
@@ -61,6 +61,7 @@ def fc(input,
 def embedding(input,
               size,
               data_type='float32',
+              is_sparse=False,
               param_attr=None,
               program=None,
               init_program=None):
@@ -72,7 +73,8 @@ def embedding(input,
         type='lookup_table',
         inputs={'Ids': input,
                 'W': w},
-        outputs={'Out': tmp})
+        outputs={'Out': tmp},
+        attrs={'is_sparse': is_sparse})
     return tmp
 
 
@@ -161,6 +163,18 @@ def func(**kwargs):
 _create_op_func_('dropout')
 
 
+def cast(x, data_type, program=None):
+    helper = LayerHelper('cast', **locals())
+    out = helper.create_tmp_variable(dtype=data_type)
+    helper.append_op(
+        type='cast',
+        inputs={'X': [x]},
+        outputs={'Out': [out]},
+        attrs={'in_data_type': x.data_type,
+               'out_data_type': out.data_type})
+    return out
+
+
 def concat(input, axis, program=None, init_program=None):
     helper = LayerHelper('concat', **locals())
     if not isinstance(input, list) and not isinstance(input, tuple):
@@ -204,6 +218,47 @@ def square_error_cost(input, label, **kwargs):
     return square_out
 
 
+def conv1d(input,
+           num_filters,
+           name=None,
+           filter_size=3,
+           act=None,
+           stride=1,
+           padding=None,
+           bias_attr=None,
+           param_attr=None,
+           program=None,
+           init_program=None):
+    # FIXME(dzh) : want to unify the argument of python layer
+    # function. So we ignore some unecessary attributes.
+    # such as, padding_trainable, context_start.
+
+    helper = LayerHelper('sequence_conv', **locals())
+    dtype = helper.input_dtype()
+
+    filter_shape = [num_filters, filter_size]
+    filter = helper.create_parameter(
+        attr=helper.param_attr, shape=filter_shape, dtype=dtype)
+    pre_bias = helper.create_tmp_variable(dtype)
+
+    input.set_lod(lod)
+    helper.append_op(
+        type='sequence_conv',
+        inputs={
+            'X': input,
+            'Filter': filter,
+        },
+        outputs={"Out": pre_bias},
+        attrs={
+            'context_stride': stride,
+            'context_start': 0,
+            'context_length': filter_size
+        })
+
+    pre_act = helper.append_bias_op(pre_bias)
+    return helper.append_activation(pre_act)
+
+
 def conv2d(input,
            num_filters,
            name=None,
@@ -256,6 +311,35 @@ def conv2d(input,
     return helper.append_activation(pre_act)
 
 
+def pool1d(input,
+           pool_size,
+           pool_type,
+           pool_stride=1,
+           pool_padding=0,
+           global_pooling=False,
+           program=None,
+           init_program=None):
+    # FIXME(dzh) : want to unify the argument of python layer
+    # function. So we ignore some unecessary attributes
+
+    ENUM_POOL_TYPE = ["max", "avg", "sqrt", "last", "first"]
+    if pool_type not in ENUM_POOL_TYPE:
+        raise ValueError("Unknown pool_type: '%s'. It can only be %s.",
+                         str(pool_type), " ".join(ENUM_POOL_TYPE))
+
+    helper = LayerHelper('sequence_pool', **locals())
+    dtype = helper.input_dtype()
+    pool_out = helper.create_tmp_variable(dtype)
+
+    helper.append_op(
+        type="sequence_pool",
+        inputs={"X": input},
+        outputs={"Out": pool_out},
+        attrs={"strategy": pool_type, })
+
+    return pool_out
+
+
 def pool2d(input,
            pool_size,
            pool_type,
@@ -275,7 +359,7 @@ def pool2d(input,
     if isinstance(pool_padding, int):
         pool_padding = [pool_padding, pool_padding]
 
-    helper = LayerHelper('conv2d', **locals())
+    helper = LayerHelper('pool2d', **locals())
     dtype = helper.input_dtype()
     pool_out = helper.create_tmp_variable(dtype)
 

python/paddle/v2/framework/nets.py

@@ -1,9 +1,11 @@
 import paddle.v2.framework.layers as layers
 
+__all__ = ["simple_img_conv_pool", "sequence_conv_pool"]
+
 
 def simple_img_conv_pool(input,
-                         filter_size,
                          num_filters,
+                         filter_size,
                          pool_size,
                          pool_stride,
                          act,
@@ -25,3 +27,29 @@ def simple_img_conv_pool(input,
         program=program,
         init_program=init_program)
     return pool_out
+
+
+def sequence_conv_pool(input,
+                       num_filters,
+                       filter_size,
+                       pool_size,
+                       pool_stride,
+                       act,
+                       program=None,
+                       init_program=None):
+    conv_out = layers.conv1d(
+        input=input,
+        num_filters=num_filters,
+        filter_size=filter_size,
+        act=act,
+        program=program,
+        init_program=init_program)
+
+    pool_out = layers.pool1d(
+        input=conv_out,
+        pool_size=pool_size,
+        pool_type='max',
+        pool_stride=pool_stride,
+        program=program,
+        init_program=init_program)
+    return pool_out