Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
afeixing77
Paddle
提交
0311fd15
P
Paddle
项目概览
afeixing77
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
0
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
未验证
提交
0311fd15
编写于
1月 29, 2018
作者:
Y
Yibing Liu
提交者:
GitHub
1月 29, 2018
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #7888 from kuke/add_lstmp_doc
Add python api for lstmp operator
上级
cce18c9f
54c160aa
变更
4
隐藏空白更改
内联
并排
Showing
4 changed file
with
217 addition
and
8 deletion
+217
-8
doc/api/v2/fluid/layers.rst
doc/api/v2/fluid/layers.rst
+5
-0
paddle/operators/CMakeLists.txt
paddle/operators/CMakeLists.txt
+1
-0
python/paddle/v2/fluid/layers/nn.py
python/paddle/v2/fluid/layers/nn.py
+199
-8
python/paddle/v2/fluid/tests/test_layers.py
python/paddle/v2/fluid/tests/test_layers.py
+12
-0
未找到文件。
doc/api/v2/fluid/layers.rst
浏览文件 @
0311fd15
...
...
@@ -18,6 +18,11 @@ dynamic_lstm
.. autofunction:: paddle.v2.fluid.layers.dynamic_lstm
:noindex:
dynamic_lstmp
-------------
.. autofunction:: paddle.v2.fluid.layers.dynamic_lstmp
:noindex:
dynamic_gru
-----------
.. autofunction:: paddle.v2.fluid.layers.dynamic_gru
...
...
paddle/operators/CMakeLists.txt
浏览文件 @
0311fd15
...
...
@@ -147,6 +147,7 @@ op_library(max_sequence_len_op DEPS lod_rank_table)
op_library
(
sequence_conv_op DEPS context_project
)
op_library
(
sequence_pool_op DEPS sequence_pooling
)
op_library
(
lstm_op DEPS sequence2batch lstm_compute
)
op_library
(
lstmp_op DEPS sequence2batch lstm_compute
)
op_library
(
gru_op DEPS sequence2batch gru_compute
)
op_library
(
recurrent_op DEPS executor
)
op_library
(
warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale math_function
)
...
...
python/paddle/v2/fluid/layers/nn.py
浏览文件 @
0311fd15
...
...
@@ -26,6 +26,7 @@ __all__ = [
'fc'
,
'embedding'
,
'dynamic_lstm'
,
'dynamic_lstmp'
,
'dynamic_gru'
,
'gru_unit'
,
'linear_chain_crf'
,
...
...
@@ -256,7 +257,8 @@ def dynamic_lstm(input,
gate_activation
=
'sigmoid'
,
cell_activation
=
'tanh'
,
candidate_activation
=
'tanh'
,
dtype
=
'float32'
):
dtype
=
'float32'
,
name
=
None
):
"""
**Dynamic LSTM Layer**
...
...
@@ -282,7 +284,7 @@ def dynamic_lstm(input,
W_{fc}, W_{oc}` are diagonal weight matrices for peephole connections. In
our implementation, we use vectors to reprenset these diagonal weight
matrices. The :math:`b` terms denote bias vectors (:math:`b_i` is the input
gate bias vector), :math:`\sigma` is the non-line activations, such as
gate bias vector), :math:`\sigma` is the non-line
ar
activations, such as
logistic sigmoid function, and :math:`i, f, o` and :math:`c` are the input
gate, forget gate, output gate, and cell activation vectors, respectively,
all of which have the same size as the cell output activation vector :math:`h`.
...
...
@@ -308,25 +310,25 @@ def dynamic_lstm(input,
(T X 4D), where T is the total time steps in this
mini-batch, D is the hidden size.
size(int): 4 * hidden size.
param_attr(ParamAttr): The parameter attribute for the learnable
param_attr(ParamAttr
|None
): The parameter attribute for the learnable
hidden-hidden weights.
- The shape is (D x 4D), where D is the hidden
size.
- Weights = {:math:`W_{ch}, W_{ih},
\
W_{fh}, W_{oh}`}
bias_attr(ParamAttr): The bias attribute for the learnable bias
- The shape is (D x 4D), where D is the hidden
size.
bias_attr(ParamAttr|None): The bias attribute for the learnable bias
weights, which contains two parts, input-hidden
bias weights and peephole connections weights if
setting `use_peepholes` to `True`.
1. `use_peepholes = False`
- The shape is (1 x 4D).
- Biases = {:math:`b_c, b_i, b_f, b_o`}.
- The shape is (1 x 4D).
2. `use_peepholes = True`
- The shape is (1 x 7D).
- Biases = { :math:`b_c, b_i, b_f, b_o, W_{ic},
\
W_{fc}, W_{oc}`}.
- The shape is (1 x 7D).
use_peepholes(bool): Whether to enable diagonal/peephole connections,
default `True`.
is_reverse(bool): Whether to compute reversed LSTM, default `False`.
...
...
@@ -339,6 +341,8 @@ def dynamic_lstm(input,
Choices = ["sigmoid", "tanh", "relu", "identity"],
default "tanh".
dtype(str): Data type. Choices = ["float32", "float64"], default "float32".
name(str|None): A name for this layer(optional). If set None, the layer
will be named automatically.
Returns:
tuple: The hidden state, and cell state of LSTM. The shape of both
\
...
...
@@ -353,6 +357,7 @@ def dynamic_lstm(input,
forward, _ = fluid.layers.dynamic_lstm(
input=forward_proj, size=hidden_dim * 4, use_peepholes=False)
"""
helper
=
LayerHelper
(
'lstm'
,
**
locals
())
size
=
size
/
4
weight
=
helper
.
create_parameter
(
...
...
@@ -389,6 +394,192 @@ def dynamic_lstm(input,
return
hidden
,
cell
def
dynamic_lstmp
(
input
,
size
,
proj_size
,
param_attr
=
None
,
bias_attr
=
None
,
use_peepholes
=
True
,
is_reverse
=
False
,
gate_activation
=
'sigmoid'
,
cell_activation
=
'tanh'
,
candidate_activation
=
'tanh'
,
proj_activation
=
'tanh'
,
dtype
=
'float32'
,
name
=
None
):
"""
**Dynamic LSTMP Layer**
LSTMP (LSTM with recurrent projection) layer has a separate projection
layer after the LSTM layer, projecting the original hidden state to a
lower-dimensional one, which is proposed to reduce the number of total
parameters and furthermore computational complexity for the LSTM,
espeacially for the case that the size of output units is relative
large (https://research.google.com/pubs/archive/43905.pdf).
The formula is as follows:
.. math::
i_t & = \sigma(W_{ix}x_{t} + W_{ir}r_{t-1} + W_{ic}c_{t-1} + b_i)
f_t & = \sigma(W_{fx}x_{t} + W_{fr}r_{t-1} + W_{fc}c_{t-1} + b_f)
\\
tilde{c_t} & = act_g(W_{cx}x_t + W_{cr}r_{t-1} + b_c)
o_t & = \sigma(W_{ox}x_{t} + W_{or}r_{t-1} + W_{oc}c_t + b_o)
c_t & = f_t \odot c_{t-1} + i_t \odot
\\
tilde{c_t}
h_t & = o_t \odot act_h(c_t)
r_t & = \overline{act_h}(W_{rh}h_t)
In the above formula:
* :math:`W`: Denotes weight matrices (e.g. :math:`W_{xi}` is
\
the matrix of weights from the input gate to the input).
* :math:`W_{ic}`, :math:`W_{fc}`, :math:`W_{oc}`: Diagonal weight
\
matrices for peephole connections. In our implementation,
\
we use vectors to reprenset these diagonal weight matrices.
* :math:`b`: Denotes bias vectors (e.g. :math:`b_i` is the input gate
\
bias vector).
* :math:`\sigma`: The activation, such as logistic sigmoid function.
* :math:`i, f, o` and :math:`c`: The input gate, forget gate, output
\
gate, and cell activation vectors, respectively, all of which have
\
the same size as the cell output activation vector :math:`h`.
* :math:`h`: The hidden state.
* :math:`r`: The recurrent projection of the hidden state.
* :math:`
\\
tilde{c_t}`: The candidate hidden state, whose
\
computation is based on the current input and previous hidden state.
* :math:`\odot`: The element-wise product of the vectors.
* :math:`act_g` and :math:`act_h`: The cell input and cell output
\
activation functions and `tanh` is usually used for them.
* :math:`\overline{act_h}`: The activation function for the projection
\
output, usually using `identity` or same as :math:`act_h`.
Set `use_peepholes` to `False` to disable peephole connection. The formula
is omitted here, please refer to the paper
http://www.bioinf.jku.at/publications/older/2604.pdf for details.
Note that these :math:`W_{xi}x_{t}, W_{xf}x_{t}, W_{xc}x_{t}, W_{xo}x_{t}`
operations on the input :math:`x_{t}` are NOT included in this operator.
Users can choose to use fully-connected layer before LSTMP layer.
Args:
input(Variable): The input of dynamic_lstmp layer, which supports
variable-time length input sequence. The underlying
tensor in this Variable is a matrix with shape
(T X 4D), where T is the total time steps in this
mini-batch, D is the hidden size.
size(int): 4 * hidden size.
proj_size(int): The size of projection output.
param_attr(ParamAttr|None): The parameter attribute for the learnable
hidden-hidden weight and projection weight.
- Hidden-hidden weight = {:math:`W_{ch}, W_{ih},
\
W_{fh}, W_{oh}`}.
- The shape of hidden-hidden weight is (P x 4D),
where P is the projection size and D the hidden
size.
- Projection weight = {:math:`W_{rh}`}.
- The shape of projection weight is (D x P).
bias_attr(ParamAttr|None): The bias attribute for the learnable bias
weights, which contains two parts, input-hidden
bias weights and peephole connections weights if
setting `use_peepholes` to `True`.
1. `use_peepholes = False`
- Biases = {:math:`b_c, b_i, b_f, b_o`}.
- The shape is (1 x 4D).
2. `use_peepholes = True`
- Biases = { :math:`b_c, b_i, b_f, b_o, W_{ic},
\
W_{fc}, W_{oc}`}.
- The shape is (1 x 7D).
use_peepholes(bool): Whether to enable diagonal/peephole connections,
default `True`.
is_reverse(bool): Whether to compute reversed LSTM, default `False`.
gate_activation(str): The activation for input gate, forget gate and
output gate. Choices = ["sigmoid", "tanh", "relu",
"identity"], default "sigmoid".
cell_activation(str): The activation for cell output. Choices = ["sigmoid",
"tanh", "relu", "identity"], default "tanh".
candidate_activation(str): The activation for candidate hidden state.
Choices = ["sigmoid", "tanh", "relu", "identity"],
default "tanh".
proj_activation(str): The activation for projection output.
Choices = ["sigmoid", "tanh", "relu", "identity"],
default "tanh".
dtype(str): Data type. Choices = ["float32", "float64"], default "float32".
name(str|None): A name for this layer(optional). If set None, the layer
will be named automatically.
Returns:
tuple: The projection of hidden state, and cell state of LSTMP. The
\
shape of projection is (T x P), for the cell state which is
\
(T x D), and both LoD is the same with the `input`.
Examples:
.. code-block:: python
hidden_dim, proj_dim = 512, 256
fc_out = fluid.layers.fc(input=input_seq, size=hidden_dim * 4,
act=None, bias_attr=None)
proj_out, _ = fluid.layers.dynamic_lstmp(input=fc_out,
size=hidden_dim * 4,
proj_size=proj_dim,
use_peepholes=False,
is_reverse=True,
cell_activation="tanh",
proj_activation="tanh")
"""
helper
=
LayerHelper
(
'lstmp'
,
**
locals
())
size
=
size
/
4
weight
=
helper
.
create_parameter
(
attr
=
helper
.
param_attr
,
shape
=
[
proj_size
,
4
*
size
],
dtype
=
dtype
)
proj_weight
=
helper
.
create_parameter
(
attr
=
helper
.
param_attr
,
shape
=
[
size
,
proj_size
],
dtype
=
dtype
)
bias_size
=
[
1
,
7
*
size
]
if
not
use_peepholes
:
bias_size
[
1
]
=
4
*
size
bias
=
helper
.
create_parameter
(
attr
=
helper
.
bias_attr
,
shape
=
bias_size
,
dtype
=
dtype
,
is_bias
=
True
)
projection
=
helper
.
create_tmp_variable
(
dtype
)
cell
=
helper
.
create_tmp_variable
(
dtype
)
ordered_proj0
=
helper
.
create_tmp_variable
(
dtype
)
batch_hidden
=
helper
.
create_tmp_variable
(
dtype
)
batch_gate
=
helper
.
create_tmp_variable
(
dtype
)
batch_cell_pre_act
=
helper
.
create_tmp_variable
(
dtype
)
helper
.
append_op
(
type
=
'lstmp'
,
inputs
=
{
'Input'
:
input
,
'Weight'
:
weight
,
'ProjWeight'
:
proj_weight
,
'Bias'
:
bias
},
outputs
=
{
'Projection'
:
projection
,
'Cell'
:
cell
,
'OrderedP0'
:
ordered_proj0
,
'BatchHidden'
:
batch_hidden
,
'BatchGate'
:
batch_gate
,
'BatchCellPreAct'
:
batch_cell_pre_act
},
attrs
=
{
'use_peepholes'
:
use_peepholes
,
'is_reverse'
:
is_reverse
,
'gate_activation'
:
gate_activation
,
'cell_activation'
:
cell_activation
,
'candidate_activation'
:
candidate_activation
,
'proj_activation'
:
proj_activation
})
return
projection
,
cell
def
dynamic_gru
(
input
,
size
,
param_attr
=
None
,
...
...
python/paddle/v2/fluid/tests/test_layers.py
浏览文件 @
0311fd15
...
...
@@ -202,6 +202,18 @@ class TestBook(unittest.TestCase):
x_t
=
x_t
,
hidden_t_prev
=
prev_hidden
,
cell_t_prev
=
prev_cell
))
print
(
str
(
program
))
def
test_dynamic_lstmp
(
self
):
program
=
Program
()
with
program_guard
(
program
):
hidden_dim
,
proj_dim
=
16
,
8
seq_data
=
layers
.
data
(
name
=
'seq_data'
,
shape
=
[
10
,
10
],
dtype
=
'float32'
,
lod_level
=
1
)
fc_out
=
layers
.
fc
(
input
=
seq_data
,
size
=
4
*
hidden_dim
)
self
.
assertIsNotNone
(
layers
.
dynamic_lstmp
(
input
=
fc_out
,
size
=
4
*
hidden_dim
,
proj_size
=
proj_dim
))
print
(
str
(
program
))
def
test_sequence_softmax
(
self
):
program
=
Program
()
with
program_guard
(
program
):
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录