m53297601
/
Paddle
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Add python wrapper for sequence_reshape.

Browse Source
add_depthwiseConv_op_gpu
yangyaming 7 years ago
parent 7ab67e32a6
commit 811c4ee401
3 changed files with 104 additions and 37 deletions
Show Stats Download Patch File Download Diff File

5
doc/api/v2/fluid/layers.rst
Unescape View File

@ -504,3 +504,8 @@ l2_normalize
------------
.. autofunction:: paddle.v2.fluid.layers.l2_normalize
:noindex:
sequence_reshape
----------------
.. autofunction:: paddle.v2.fluid.layers.sequence_reshape
:noindex:

128
python/paddle/v2/fluid/layers/nn.py
Unescape View File

@ -28,7 +28,7 @@ __all__ = [
'batch_norm', 'beam_search_decode', 'conv2d_transpose', 'sequence_expand',
'lstm_unit', 'reduce_sum', 'reduce_mean', 'reduce_max', 'reduce_min',
'sequence_first_step', 'sequence_last_step', 'dropout', 'split',
'l2_normalize', 'matmul', 'warpctc'
'l2_normalize', 'matmul', 'warpctc', 'sequence_reshape'
]
@ -213,33 +213,33 @@ def dynamic_lstm(input,
(https://arxiv.org/pdf/1402.1128.pdf), the formula is as follows:
.. math::
i_t & = \sigma(W_{ix}x_{t} + W_{ih}h_{t-1} + W_{ic}c_{t-1} + b_i)
f_t & = \sigma(W_{fx}x_{t} + W_{fh}h_{t-1} + W_{fc}c_{t-1} + b_f)
i_t & = \sigma(W_{ix}x_{t} + W_{ih}h_{t-1} + W_{ic}c_{t-1} + b_i)
\\tilde{c_t} & = act_g(W_{cx}x_t + W_{ch}h_{t-1} + b_c)
f_t & = \sigma(W_{fx}x_{t} + W_{fh}h_{t-1} + W_{fc}c_{t-1} + b_f)
o_t & = \sigma(W_{ox}x_{t} + W_{oh}h_{t-1} + W_{oc}c_t + b_o)
\\tilde{c_t} & = act_g(W_{cx}x_t + W_{ch}h_{t-1} + b_c)
c_t & = f_t \odot c_{t-1} + i_t \odot \\tilde{c_t}
o_t & = \sigma(W_{ox}x_{t} + W_{oh}h_{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)
where the :math:`W` terms denote weight matrices (e.g. :math:`W_{xi}` is
where the :math:`W` terms denote weight matrices (e.g. :math:`W_{xi}` is
the matrix of weights from the input gate to the input), :math:`W_{ic}, \
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
logistic sigmoid function, and :math:`i, f, o` and :math:`c` are the input
gate, forget gate, output gate, and cell activation vectors, respectively,
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
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`.
The :math:`\odot` is the element-wise product of the vectors. :math:`act_g`
and :math:`act_h` are the cell input and cell output activation functions
and `tanh` is usually used for them. :math:`\\tilde{c_t}` is also called
candidate hidden state, which is computed based on the current input and
The :math:`\odot` is the element-wise product of the vectors. :math:`act_g`
and :math:`act_h` are the cell input and cell output activation functions
and `tanh` is usually used for them. :math:`\\tilde{c_t}` is also called
candidate hidden state, which is computed based on the current input and
the previous hidden state.
Set `use_peepholes` to `False` to disable peephole connection. The formula
@ -251,38 +251,38 @@ def dynamic_lstm(input,
Users can choose to use fully-connect layer before LSTM layer.
Args:
input(Variable): The input of dynamic_lstm 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
input(Variable): The input of dynamic_lstm 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.
param_attr(ParamAttr): The parameter attribute for the learnable
hidden-hidden weights.
param_attr(ParamAttr): The parameter attribute for the learnable
hidden-hidden weights.
- The shape is (D x 4D), where D is the hidden
size.
- 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
weights, which contains two parts, input-hidden
bias weights and peephole connections weights if
setting `use_peepholes` to `True`.
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).
1. `use_peepholes = False`
- The shape is (1 x 4D).
- Biases = {:math:`b_c, b_i, b_f, b_o`}.
2. `use_peepholes = True`
- The shape is (1 x 7D).
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}`}.
use_peepholes(bool): Whether to enable diagonal/peephole connections,
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",
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",
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"],
@ -1914,3 +1914,57 @@ def warpctc(input, label, blank=0, norm_by_times=False, **kwargs):
attrs={'blank': blank,
'norm_by_times': norm_by_times})
return loss_out
def sequence_reshape(input, new_dim):
"""
**Sequence Reshape Layer**
This layer will rearrange the input sequences. The new dimension is set by
user. Length of each sequence is computed according to original length,
original dimension and new dimension. The following example will help to
illustrate the function of this layer:
.. code-block:: text
x is a LoDTensor:
x.lod = [[0, 2, 6]]
x.data = [[1, 2], [3, 4],
[5, 6], [7, 8], [9, 10], [11, 12]]
x.dims = [6, 2]
set new_dim = 4
then out is a LoDTensor:
out.lod = [[0, 1, 3]]
out.data = [[1, 2, 3, 4],
[5, 6, 7, 8], [9, 10, 11, 12]]
out.dims = [3, 4]
Currently, only 1-level LoDTensor is supported and please make sure
(original length * original dimension) can be divided by new dimension with
no remainder for each sequence.
Args:
input (Variable): (LodTensor, default: LoDTensor<float>), a 2-D LoDTensor
with shape being [N, M] where M for dimension.
new_dim (int): New dimension which the input LoDTensor is reshaped to.
Returns:
Variable: Reshaped LoDTensor according to new dimension.
Examples:
.. code-block:: python
x = fluid.layers.data(name='x', shape=[5, 20],
dtype='float32', lod_level=1)
x_reshaped = layers.sequence_reshape(input=x, new_dim=10)
"""
helper = LayerHelper('sequence_reshape', **locals())
out = helper.create_tmp_variable(helper.input_dtype())
helper.append_op(
type='sequence_reshape',
inputs={'X': [input]},
outputs={'Out': [out]},
attrs={'new_dim': new_dim})
return out

8
python/paddle/v2/fluid/tests/test_layers.py
Unescape View File

@ -216,6 +216,14 @@ class TestBook(unittest.TestCase):
self.assertIsNotNone(x)
print(str(program))
def test_sequence_reshape(self):
program = Program()
with program_guard(program):
x = layers.data(name='x', shape=[8], dtype='float32', lod_level=1)
out = layers.sequence_reshape(input=x, new_dim=16)
self.assertIsNotNone(out)
print(str(program))
if __name__ == '__main__':
unittest.main()

Write Preview
Loading…
Cancel
Save