rnn_config_en.rst 14.0 KB
Newer Older
1 2
RNN Configuration
=================
3

W
wangjiang03 已提交
4
This tutorial will guide you how to configure recurrent neural network in PaddlePaddle. PaddlePaddle supports highly flexible and efficient recurrent neural network configuration. In this tutorial, you will learn how to:
5

W
wangjiang03 已提交
6 7 8
- configure recurrent neural network architecture.
- generate sequence with learned recurrent neural network models.

L
Luo Tao 已提交
9 10
We will use vanilla recurrent neural network, and sequence to sequence model to guide you through these steps. The code of sequence to sequence model can be found at `book/08.machine_translation <https://github.com/PaddlePaddle/book/tree/develop/08.machine_translation>`_ .
And the data preparation of this model can be found at `python/paddle/v2/dataset/wmt14.py <https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/dataset/wmt14.py>`_ 
W
wangjiang03 已提交
11

12
===============================================
W
wangjiang03 已提交
13
Configure Recurrent Neural Network Architecture
14
===============================================
W
wangjiang03 已提交
15

16
-------------------------------------
W
wangjiang03 已提交
17
Simple Gated Recurrent Neural Network
18 19
-------------------------------------

W
wangjiang03 已提交
20 21
Recurrent neural network process a sequence at each time step sequentially. An example of the architecture of LSTM is listed below.

22
.. image:: src/bi_lstm.jpg
L
liaogang 已提交
23
     :align: center
W
wangjiang03 已提交
24

25
Generally speaking, a recurrent network perform the following operations from :math:`t=1` to :math:`t=T`, or reversely from :math:`t=T` to :math:`t=1`.
W
wangjiang03 已提交
26 27 28 29 30 31

.. math::

    x_{t+1} = f_x(x_t), y_t = f_y(x_t)


32
where :math:`f_x(.)` is called **step function**, and :math:`f_y(.)` is called **output function**. In vanilla recurrent neural network, both of the step function and output function are very simple. However, PaddlePaddle supports the configuration of very complex architectures by modifying these two functions. We will use the sequence to sequence model with attention as an example to demonstrate how you can configure complex recurrent neural network models. In this section, we will use a simple vanilla recurrent neural network as an example of configuring simple recurrent neural network using :code:`recurrent_group`. Notice that if you only need to use simple RNN, GRU, or LSTM, then :code:`grumemory` and :code:`lstmemory` is recommended because they are more computationally efficient than :code:`recurrent_group`.
W
wangjiang03 已提交
33

34
For vanilla RNN, at each time step, the **step function** is:
W
wangjiang03 已提交
35 36 37 38 39 40

.. math::

    x_{t+1} = W_x x_t + W_i I_t + b

where :math:`x_t` is the RNN state, and :math:`I_t` is the input, :math:`W_x` and :math:`W_i` are transformation matrices for RNN states and inputs, respectively. :math:`b` is the bias.
41
Its **output function** simply takes :math:`x_t` as the output.
W
wangjiang03 已提交
42

43
:code:`recurrent_group` is the most important tools for constructing recurrent neural networks. It defines the **step function**, **output function** and the inputs of the recurrent neural network. Notice that the :code:`step` argument of this function implements both the :code:`step function` and the :code:`output function`:
W
wangjiang03 已提交
44

45
.. code-block:: python
W
wangjiang03 已提交
46 47 48 49 50 51 52 53 54

    def simple_rnn(input,
                   size=None,
                   name=None,
                   reverse=False,
                   rnn_bias_attr=None,
                   act=None,
                   rnn_layer_attr=None):
        def __rnn_step__(ipt):
L
Luo Tao 已提交
55 56 57 58 59 60 61 62
           out_mem = paddle.layer.memory(name=name, size=size)
           rnn_out = paddle.layer.mixed(input = [paddle.layer.full_matrix_projection(input=ipt),
                                                 paddle.layer.full_matrix_projection(input=out_mem)],
                                        name = name,
                                        bias_attr = rnn_bias_attr,
                                        act = act,
                                        layer_attr = rnn_layer_attr,
                                        size = size)
W
wangjiang03 已提交
63
           return rnn_out
L
Luo Tao 已提交
64 65 66 67
        return paddle.layer.recurrent_group(name='%s_recurrent_group' % name,
                                            step=__rnn_step__,
                                            reverse=reverse,
                                            input=input)
W
wangjiang03 已提交
68 69


70
PaddlePaddle uses memory to construct step function. **Memory** is the most important concept when constructing recurrent neural networks in PaddlePaddle. A memory is a state that is used recurrently in step functions, such as :math:`x_{t+1} = f_x(x_t)`. One memory contains an **output** and a **input**. The output of memory at the current time step is utilized as the input of the memory at the next time step. A memory can also has a **boot layer**, whose output is utilized as the initial value of the memory. In our case, the output of the gated recurrent unit is employed as the output memory. Notice that the name of the layer :code:`rnn_out` is the same as the name of :code:`out_mem`. This means the output of the layer :code:`rnn_out` (:math:`x_{t+1}`) is utilized as the **output** of :code:`out_mem` memory.
W
wangjiang03 已提交
71 72 73

A memory can also be a sequence. In this case, at each time step, we have a sequence as the state of the recurrent neural network. This can be useful when constructing very complex recurrent neural network. Other advanced functions include defining multiple memories, and defining hierarchical recurrent neural network architecture using sub-sequence.

74
We return :code:`rnn_out` at the end of the function. It means that the output of the layer :code:`rnn_out` is utilized as the **output** function of the gated recurrent neural network.
W
wangjiang03 已提交
75

76
-----------------------------------------
W
wangjiang03 已提交
77
Sequence to Sequence Model with Attention
78
-----------------------------------------
W
wangjiang03 已提交
79 80
We will use the sequence to sequence model with attention as an example to demonstrate how you can configure complex recurrent neural network models. An illustration of the sequence to sequence model with attention is shown in the following figure.

81
.. image:: src/encoder-decoder-attention-model.png
L
liaogang 已提交
82
      :align: center
W
wangjiang03 已提交
83

84
In this model, the source sequence :math:`S = \{s_1, \dots, s_T\}` is encoded with a bidirectional gated recurrent neural networks. The hidden states of the bidirectional gated recurrent neural network :math:`H_S = \{H_1, \dots, H_T\}` is called *encoder vector* The decoder is a gated recurrent neural network. When decoding each token :math:`y_t`, the gated recurrent neural network generates a set of weights :math:`W_S^t = \{W_1^t, \dots, W_T^t\}`, which are used to compute a weighted sum of the encoder vector. The weighted sum of the encoder vector is utilized to condition the generation of the token :math:`y_t`.
W
wangjiang03 已提交
85

L
Luo Tao 已提交
86
The encoder part of the model is listed below. It calls :code:`grumemory` to represent gated recurrent neural network. It is the recommended way of using recurrent neural network if the network architecture is simple, because it is faster than :code:`recurrent_group`. We have implemented most of the commonly used recurrent neural network architectures, you can refer to :ref:`api_trainer_config_helpers_layers` for more details.
W
wangjiang03 已提交
87

88
We also project the encoder vector to :code:`decoder_size` dimensional space, get the first instance of the backward recurrent network, and project it to :code:`decoder_size` dimensional space:
W
wangjiang03 已提交
89

90
.. code-block:: python
W
wangjiang03 已提交
91 92

    # Define the data layer of the source sentence.
L
Luo Tao 已提交
93 94 95
    src_word_id = paddle.layer.data(
        name='source_language_word',
        type=paddle.data_type.integer_value_sequence(source_dict_dim))
W
wangjiang03 已提交
96
    # Calculate the word embedding of each word.
L
Luo Tao 已提交
97
    src_embedding = paddle.layer.embedding(
W
wangjiang03 已提交
98 99
        input=src_word_id,
        size=word_vector_dim,
L
Luo Tao 已提交
100
        param_attr=paddle.attr.ParamAttr(name='_source_language_embedding'))
W
wangjiang03 已提交
101
    # Apply forward recurrent neural network.
L
Luo Tao 已提交
102 103
    src_forward = paddle.networks.simple_gru(
        input=src_embedding, size=encoder_size)
W
wangjiang03 已提交
104
    # Apply backward recurrent neural network. reverse=True means backward recurrent neural network.
L
Luo Tao 已提交
105 106
    src_backward = paddle.networks.simple_gru(
        input=src_embedding, size=encoder_size, reverse=True)
W
wangjiang03 已提交
107
    # Mix the forward and backward parts of the recurrent neural network together.
L
Luo Tao 已提交
108
    encoded_vector = paddle.layer.concat(input=[src_forward, src_backward])
W
wangjiang03 已提交
109 110

    # Project encoding vector to decoder_size.
L
Luo Tao 已提交
111 112 113
    encoded_proj = paddle.layer.mixed(
        size=decoder_size,
        input=paddle.layer.full_matrix_projection(encoded_vector))
W
wangjiang03 已提交
114 115

    # Compute the first instance of the backward RNN.
L
Luo Tao 已提交
116
    backward_first = paddle.layer.first_seq(input=src_backward)
W
wangjiang03 已提交
117 118

    # Project the first instance of backward RNN to decoder size.
L
Luo Tao 已提交
119 120 121 122
    decoder_boot = paddle.layer.mixed(
       size=decoder_size,
       act=paddle.activation.Tanh(),
       input=paddle.layer.full_matrix_projection(backward_first))
W
wangjiang03 已提交
123 124


125
The decoder uses :code:`recurrent_group` to define the recurrent neural network. The step and output functions are defined in :code:`gru_decoder_with_attention`:
W
wangjiang03 已提交
126

127
.. code-block:: python
128

L
Luo Tao 已提交
129 130 131 132 133 134 135 136 137 138
    group_input1 = paddle.layer.StaticInput(input=encoded_vector, is_seq=True)
    group_input2 = paddle.layer.StaticInput(input=encoded_proj, is_seq=True)
    group_inputs = [group_input1, group_input2]
    trg_embedding = paddle.layer.embedding(
            input=paddle.layer.data(
                name='target_language_word',
                type=paddle.data_type.integer_value_sequence(target_dict_dim)),
            size=word_vector_dim,
            param_attr=paddle.attr.ParamAttr(name='_target_language_embedding'))
        group_inputs.append(trg_embedding)
139 140
    group_inputs.append(trg_embedding)

W
wangjiang03 已提交
141 142 143 144 145 146
    # For decoder equipped with attention mechanism, in training,
    # target embedding (the groudtruth) is the data input,
    # while encoded source sequence is accessed to as an unbounded memory.
    # StaticInput means the same value is utilized at different time steps.
    # Otherwise, it is a sequence input. Inputs at different time steps are different.
    # All sequence inputs should have the same length.
L
Luo Tao 已提交
147 148 149 150
    decoder = paddle.layer.recurrent_group(
            name=decoder_group_name,
            step=gru_decoder_with_attention,
            input=group_inputs)
W
wangjiang03 已提交
151 152


153
The implementation of the step function is listed as below. First, it defines the **memory** of the decoder network. Then it defines attention, gated recurrent unit step function, and the output function:
W
wangjiang03 已提交
154

155
.. code-block:: python
W
wangjiang03 已提交
156 157 158 159 160

    def gru_decoder_with_attention(enc_vec, enc_proj, current_word):
        # Defines the memory of the decoder.
        # The output of this memory is defined in gru_step.
        # Notice that the name of gru_step should be the same as the name of this memory.
L
Luo Tao 已提交
161 162
        decoder_mem = paddle.layer.memory(
            name='gru_decoder', size=decoder_size, boot_layer=decoder_boot)
W
wangjiang03 已提交
163
        # Compute attention weighted encoder vector.
L
Luo Tao 已提交
164 165 166 167
        context = paddle.networks.simple_attention(
            encoded_sequence=enc_vec,
            encoded_proj=enc_proj,
            decoder_state=decoder_mem)
W
wangjiang03 已提交
168
        # Mix the current word embedding and the attention weighted encoder vector.
L
Luo Tao 已提交
169 170 171 172 173 174
        decoder_inputs = paddle.layer.mixed(
            size=decoder_size * 3,
            input=[
                paddle.layer.full_matrix_projection(input=context),
                paddle.layer.full_matrix_projection(input=current_word)
            ])
W
wangjiang03 已提交
175
        # Define Gated recurrent unit recurrent neural network step function.
L
Luo Tao 已提交
176 177 178 179 180
        gru_step = paddle.layer.gru_step(
            name='gru_decoder',
            input=decoder_inputs,
            output_mem=decoder_mem,
            size=decoder_size)
W
wangjiang03 已提交
181
        # Defines the output function.
L
Luo Tao 已提交
182 183 184 185 186
        out = paddle.layer.mixed(
            size=target_dict_dim,
            bias_attr=True,
            act=paddle.activation.Softmax(),
            input=paddle.layer.full_matrix_projection(input=gru_step))
W
wangjiang03 已提交
187 188 189 190 191 192
        return out


=================
Generate Sequence
=================
193
After training the model, we can use it to generate sequences. A common practice is to use **beam search** to generate sequences. The following code snippets defines a beam search algorithm. Notice that :code:`beam_search` function assumes the output function of the :code:`step` returns a softmax normalized probability vector of the next token. We made the following changes to the model.
W
wangjiang03 已提交
194

195 196
* use :code:`GeneratedInput` for trg_embedding. :code:`GeneratedInput` computes the embedding of the generated token at the last time step for the input at the current time step.
* use :code:`beam_search` function. This function needs to set:
W
wangjiang03 已提交
197

198 199 200 201
  - :code:`bos_id`: the start token. Every sentence starts with the start token.
  - :code:`eos_id`: the end token. Every sentence ends with the end token.
  - :code:`beam_size`: the beam size used in beam search.
  - :code:`max_length`: the maximum length of the generated sentences.
202
    
203
The code is listed below:
W
wangjiang03 已提交
204

205
.. code-block:: python
W
wangjiang03 已提交
206

L
Luo Tao 已提交
207 208 209
    group_input1 = paddle.layer.StaticInput(input=encoded_vector, is_seq=True)
    group_input2 = paddle.layer.StaticInput(input=encoded_proj, is_seq=True)
    group_inputs = [group_input1, group_input2]
W
wangjiang03 已提交
210 211 212 213 214 215
    # In generation, decoder predicts a next target word based on
    # the encoded source sequence and the last generated target word.
    # The encoded source sequence (encoder's output) must be specified by
    # StaticInput which is a read-only memory.
    # Here, GeneratedInputs automatically fetchs the last generated word,
    # which is initialized by a start mark, such as <s>.
L
Luo Tao 已提交
216 217 218 219
    trg_embedding = paddle.layer.GeneratedInput(
            size=target_dict_dim,
            embedding_name='_target_language_embedding',
            embedding_size=word_vector_dim)
220
    group_inputs.append(trg_embedding)
L
Luo Tao 已提交
221 222 223 224 225 226 227 228
    beam_gen = paddle.layer.beam_search(
            name=decoder_group_name,
            step=gru_decoder_with_attention,
            input=group_inputs,
            bos_id=0, # Beginnning token.
            eos_id=1, # End of sentence token.
            beam_size=beam_size,
            max_length=max_length)
229

L
Luo Tao 已提交
230
    return beam_gen
W
wangjiang03 已提交
231 232


L
Luo Tao 已提交
233
Notice that this generation technique is only useful for decoder like generation process. If you are working on sequence tagging tasks, please refer to `book/06.understand_sentiment <https://github.com/PaddlePaddle/book/tree/develop/06.understand_sentiment>`_ for more details.
W
wangjiang03 已提交
234

L
Luo Tao 已提交
235
The full configuration file is located at `book/08.machine_translation/train.py <https://github.com/PaddlePaddle/book/blob/develop/08.machine_translation/train.py>`_ .