提交 0fbfd2dc 编写于 作者: Y Yibing Liu

Simplify the symbol description

上级 634faab1
...@@ -435,25 +435,28 @@ def dynamic_lstmp(input, ...@@ -435,25 +435,28 @@ def dynamic_lstmp(input,
r_t & = \overline{act_h}(W_{rh}h_t) r_t & = \overline{act_h}(W_{rh}h_t)
where the :math:`W` terms denote weight matrices (e.g. :math:`W_{xi}` is In the above formula:
the matrix of weights from the input gate to the input), :math:`W_{ic}`,
:math:`W_{fc}`, :math:`W_{oc}` are diagonal weight matrices for peephole * :math:`W`: Denotes weight matrices (e.g. :math:`W_{xi}` is \
connections. In our implementation, we use vectors to reprenset these the matrix of weights from the input gate to the input).
diagonal weight matrices. The :math:`b` terms denote bias vectors * :math:`W_{ic}`, :math:`W_{fc}`, :math:`W_{oc}`: Diagonal weight \
(:math:`b_i` is the input gate bias vector), :math:`\sigma` is the matrices for peephole connections. In our implementation, \
activation, such as logistic sigmoid function, and :math:`i, f, o` and we use vectors to reprenset these diagonal weight matrices.
:math:`c` are the input gate, forget gate, output gate, and cell activation * :math:`b`: Denotes bias vectors (e.g. :math:`b_i` is the input gate \
vectors, respectively, all of which have the same size as the cell output bias vector).
activation vector :math:`h`. Here :math:`h` is usually called the hidden * :math:`\sigma`: The activation, such as logistic sigmoid function.
state and :math:`r` denotes its recurrent projection. And * :math:`i, f, o` and :math:`c`: The input gate, forget gate, output \
:math:`\\tilde{c_t}` is also called the candidate hidden state, whose gate, and cell activation vectors, respectively, all of which have \
computation is based on the current input and previous hidden state. the same size as the cell output activation vector :math:`h`.
* :math:`h`: The hidden state.
The :math:`\odot` is the element-wise product of the vectors. :math:`act_g` * :math:`r`: The recurrent projection of the hidden state.
and :math:`act_h` are the cell input and cell output activation functions * :math:`\\tilde{c_t}`: The candidate hidden state, whose \
and `tanh` is usually used for them. :math:`\overline{act_h}` is the computation is based on the current input and previous hidden state.
activation function for the projection output, usually using `identity` or * :math:`\odot`: The element-wise product of the vectors.
same as :math:`act_h`. * :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 Set `use_peepholes` to `False` to disable peephole connection. The formula
is omitted here, please refer to the paper is omitted here, please refer to the paper
...@@ -519,12 +522,16 @@ def dynamic_lstmp(input, ...@@ -519,12 +522,16 @@ def dynamic_lstmp(input,
Examples: Examples:
.. code-block:: python .. code-block:: python
hidden_dim = 512 hidden_dim, proj_dim = 512, 256
proj_dim = 256
fc_out = fluid.layers.fc(input=input_seq, size=hidden_dim * 4, fc_out = fluid.layers.fc(input=input_seq, size=hidden_dim * 4,
act=None, bias_attr=None) act=None, bias_attr=None)
proj_out, _ = fluid.layers.dynamic_lstmp(input=fc_out, proj_out, _ = fluid.layers.dynamic_lstmp(input=fc_out,
size=hidden_dim * 4, proj_size=proj_dim, use_peepholes=False) size=hidden_dim * 4,
proj_size=proj_dim,
use_peepholes=False,
is_reverse=True,
cell_activation="tanh",
proj_activation="tanh")
""" """
helper = LayerHelper('lstmp', **locals()) helper = LayerHelper('lstmp', **locals())
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册