new_layer_en.rst 19.6 KB
Newer Older
L
Luo Tao 已提交
1 2 3
================
Write New Layers
================
W
wangjiang03 已提交
4

Z
zhangjinchao01 已提交
5
This tutorial will guide you to write customized layers in PaddlePaddle. We will utilize fully connected layer as an example to guide you through the following steps for writing a new layer.
W
wangjiang03 已提交
6

Z
zhangjinchao01 已提交
7 8
- Derive equations for the forward and backward part of the layer.
- Implement C++ class for the layer.
9 10
- Write gradient check unit test to make sure the gradients are correctly computed.
- Implement Python wrapper for the layer.
Z
zhangjinchao01 已提交
11

W
wangjiang03 已提交
12
Derive Equations
13
================
W
wangjiang03 已提交
14

Z
zhangjinchao01 已提交
15 16 17
First we need to derive equations of the *forward* and *backward* part of the layer. The forward part computes the output given an input. The backward part computes the gradients of the input and the parameters given the the gradients of the output.

The illustration of a fully connected layer is shown in the following figure. In a fully connected layer, all output nodes are connected to all the input nodes.
W
wangjiang03 已提交
18

_青葱's avatar
_青葱 已提交
19
..  image:: src/FullyConnected.jpg
20 21
    :align: center
    :scale: 60 %
Z
zhangjinchao01 已提交
22 23

The *forward part* of a layer transforms an input into the corresponding output.
24
Fully connected layer takes a dense input vector with dimension :math:`D_i`. It uses a transformation matrix :math:`W` with size :math:`D_i \times D_o` to project :math:`x` into a :math:`D_o` dimensional vector, and add a bias vector :math:`b` with dimension :math:`D_o` to the vector.
W
wangjiang03 已提交
25 26 27

.. math::

28
   y = f(W^T x + b)
W
wangjiang03 已提交
29 30 31

where :math:`f(.)` is an nonlinear *activation* function, such as sigmoid, tanh, and Relu.

32
The transformation matrix :math:`W` and bias vector :math:`b` are the *parameters* of the layer. The *parameters* of a layer are learned during training in the *backward pass*. The backward pass computes the gradients of the output function with respect to all parameters and inputs. The optimizer can use chain rule to compute the gradients of the loss function with respect to each parameter.
33 34

Suppose our loss function is :math:`c(y)`, then
W
wangjiang03 已提交
35 36 37

.. math::

38
   \frac{\partial c(y)}{\partial x} = \frac{\partial c(y)}{\partial y} \frac{\partial y}{\partial x}
W
wangjiang03 已提交
39

40
Suppose :math:`z = W^T x + b`, then
W
wangjiang03 已提交
41 42 43

.. math::

44
   \frac{\partial y}{\partial z} = \frac{\partial f(z)}{\partial z}
Z
zhangjinchao01 已提交
45

W
wangjiang03 已提交
46
This derivative can be automatically computed by our base layer class.
Z
zhangjinchao01 已提交
47

48
Then, for fully connected layer, we need to compute:
Z
zhangjinchao01 已提交
49

W
wangjiang03 已提交
50
.. math::
51

52
   \frac{\partial z}{\partial x} = W, \frac{\partial z_j}{\partial W_{ij}} = x_i, \frac{\partial z}{\partial b} = \mathbf 1
Z
zhangjinchao01 已提交
53

54
where :math:`\mathbf 1` is an all one vector, :math:`W_{ij}` is the number at the i-th row and j-th column of the matrix :math:`W`, :math:`z_j` is the j-th component of the vector :math:`z`, and :math:`x_i` is the i-th component of the vector :math:`x`.
W
wangjiang03 已提交
55

56
Finally we can use chain rule to calculate :math:`\frac{\partial z}{\partial x}`, and :math:`\frac{\partial z}{\partial W}`. The details of the computation will be given in the next section.
W
wangjiang03 已提交
57 58

Implement C++ Class
59
===================
Z
zhangjinchao01 已提交
60

X
Xin Pan 已提交
61
The C++ class of the layer implements the initialization, forward, and backward part of the layer. The fully connected layer is at :code:`paddle/legacy/gserver/layers/FullyConnectedLayer.h` and :code:`paddle/legacy/gserver/layers/FullyConnectedLayer.cpp`. We list simplified version of the code below.
Z
zhangjinchao01 已提交
62

Q
qingqing01 已提交
63
It needs to derive the base class :code:`paddle::Layer`, and it needs to override the following functions:
Z
zhangjinchao01 已提交
64 65

- constructor and destructor.
66 67 68 69 70
- :code:`init` function. It is used to initialize the parameters and settings.
- :code:`forward`. It implements the forward part of the layer.
- :code:`backward`. It implements the backward part of the layer.
- :code:`prefetch`. It is utilized to determine the rows corresponding parameter matrix to prefetch from parameter server. You do not need to override this function if your layer does not need remote sparse update. (most layers do not need to support remote sparse update)

Z
zhangjinchao01 已提交
71

72
The header file is listed below:
Z
zhangjinchao01 已提交
73

74
.. code-block:: c++
Z
zhangjinchao01 已提交
75

W
wangjiang03 已提交
76 77 78 79 80 81 82 83
    namespace paddle {
    /**
     * A layer has full connections to all neurons in the previous layer.
     * It computes an inner product with a set of learned weights, and
     * (optionally) adds biases.
     *
     * The config file api is fc_layer.
     */
Z
zhangjinchao01 已提交
84

W
wangjiang03 已提交
85 86 87 88
    class FullyConnectedLayer : public Layer {
    protected:
      WeightList weights_;
      std::unique_ptr<Weight> biases_;
Z
zhangjinchao01 已提交
89

W
wangjiang03 已提交
90 91
    public:
      explicit FullyConnectedLayer(const LayerConfig& config)
92
          : Layer(config) {}
W
wangjiang03 已提交
93
      ~FullyConnectedLayer() {}
Z
zhangjinchao01 已提交
94

W
wangjiang03 已提交
95
      bool init(const LayerMap& layerMap, const ParameterMap& parameterMap);
Z
zhangjinchao01 已提交
96

W
wangjiang03 已提交
97 98 99 100 101 102 103
      Weight& getWeight(int idx) { return *weights_[idx]; }

      void prefetch();
      void forward(PassType passType);
      void backward(const UpdateCallback& callback = nullptr);
    };
    }  // namespace paddle
Z
zhangjinchao01 已提交
104

105
It defines the parameters as class variables. We use :code:`Weight` class as abstraction of parameters. It supports multi-thread update. The details of this class will be described in details in the implementations.
Z
zhangjinchao01 已提交
106

107 108
- :code:`weights_` is a list of weights for the transformation matrices. The current implementation can have more than one inputs. Thus, it has a list of weights. One weight corresponds to an input.
- :code:`biases_` is a weight for the bias vector.
Z
zhangjinchao01 已提交
109

110
The fully connected layer does not have layer configuration hyper-parameters. If there are some layer hyper-parameters, a common practice is to store it in :code:`LayerConfig& config`, and put it into a class variable in the constructor.
Z
zhangjinchao01 已提交
111

112 113 114
The following code snippet implements the :code:`init` function.

- First, every :code:`init` function must call the :code:`init` function of the base class :code:`Layer::init(layerMap, parameterMap);`. This statement will initialize the required variables and connections for each layer.
W
wangjiang03 已提交
115
- The it initializes all the weights matrices :math:`W`. The current implementation can have more than one inputs. Thus, it has a list of weights.
Z
zhangjinchao01 已提交
116 117 118
- Finally, it initializes the bias.


119
.. code-block:: c++
Z
zhangjinchao01 已提交
120

W
wangjiang03 已提交
121
    bool FullyConnectedLayer::init(const LayerMap& layerMap,
122
                                   const ParameterMap& parameterMap) {
W
wangjiang03 已提交
123 124
      /* Initialize the basic parent class */
      Layer::init(layerMap, parameterMap);
Z
zhangjinchao01 已提交
125

W
wangjiang03 已提交
126 127 128
      /* initialize the weightList */
      CHECK(inputLayers_.size() == parameters_.size());
      for (size_t i = 0; i < inputLayers_.size(); i++) {
129 130 131 132 133 134 135 136 137 138 139 140 141 142
        // Option the parameters
        size_t height = inputLayers_[i]->getSize();
        size_t width = getSize();

        // create a new weight
        if (parameters_[i]->isSparse()) {
          CHECK_LE(parameters_[i]->getSize(), width * height);
        } else {
          CHECK_EQ(parameters_[i]->getSize(), width * height);
        }
        Weight* w = new Weight(height, width, parameters_[i]);

        // append the new weight to the list
        weights_.emplace_back(w);
W
wangjiang03 已提交
143 144 145 146
      }

      /* initialize biases_ */
      if (biasParameter_.get() != NULL) {
147
        biases_ = std::unique_ptr<Weight>(new Weight(1, getSize(), biasParameter_));
W
wangjiang03 已提交
148
      }
Z
zhangjinchao01 已提交
149

W
wangjiang03 已提交
150 151
      return true;
    }
Z
zhangjinchao01 已提交
152 153 154

The implementation of the forward part has the following steps.

155 156 157 158 159 160 161
- Every layer must call :code:`Layer::forward(passType);` at the beginning of its :code:`forward` function.
- Then it allocates memory for the output using :code:`reserveOutput(batchSize, size);`. This step is necessary because we support the batches to have different batch sizes. :code:`reserveOutput` will change the size of the output accordingly. For the sake of efficiency, we will allocate new memory if we want to expand the matrix, but we will reuse the existing memory block if we want to shrink the matrix.
- Then it computes :math:`\sum_i W_i x + b` using Matrix operations. :code:`getInput(i).value` retrieve the matrix of the i-th input. Each input is a :math:`batchSize \times dim` matrix, where each row represents an single input in a batch. For a complete lists of supported matrix operations, please refer to :code:`paddle/math/Matrix.h` and :code:`paddle/math/BaseMatrix.h`.
- Finally it applies the activation function using :code:`forwardActivation();`. It will automatically applies the corresponding activation function specifies in the network configuration.


.. code-block:: c++
Z
zhangjinchao01 已提交
162

W
wangjiang03 已提交
163 164
    void FullyConnectedLayer::forward(PassType passType) {
      Layer::forward(passType);
Z
zhangjinchao01 已提交
165

W
wangjiang03 已提交
166 167 168
      /* malloc memory for the output_ if necessary */
      int batchSize = getInput(0).getBatchSize();
      int size = getSize();
Z
zhangjinchao01 已提交
169

W
wangjiang03 已提交
170
      {
171 172
        // Settup the size of the output.
        reserveOutput(batchSize, size);
W
wangjiang03 已提交
173
      }
Z
zhangjinchao01 已提交
174

W
wangjiang03 已提交
175
      MatrixPtr outV = getOutputValue();
Z
zhangjinchao01 已提交
176

W
wangjiang03 已提交
177 178
      // Apply the the transformation matrix to each input.
      for (size_t i = 0; i != inputLayers_.size(); ++i) {
179 180 181 182
        auto input = getInput(i);
        CHECK(input.value) << "The input of 'fc' layer must be matrix";
        i == 0 ? outV->mul(input.value, weights_[i]->getW(), 1, 0)
               : outV->mul(input.value, weights_[i]->getW(), 1, 1);
W
wangjiang03 已提交
183
      }
Z
zhangjinchao01 已提交
184

W
wangjiang03 已提交
185 186
      /* add the bias-vector */
      if (biases_.get() != NULL) {
187
        outV->addBias(*(biases_->getW()), 1);
W
wangjiang03 已提交
188 189 190
      }

      /* activation */ {
191
        forwardActivation();
W
wangjiang03 已提交
192 193
      }
    }
Z
zhangjinchao01 已提交
194 195

The implementation of the backward part has the following steps.
196 197 198 199

- :code:`backwardActivation()` computes the gradients of the activation. The gradients will be multiplies in place to the gradients of the output, which can be retrieved using :code:`getOutputGrad()`.
- Compute the gradients of bias. Notice that we an use :code:`biases_->getWGrad()` to get the gradient matrix of the corresponding parameter. After the gradient of one parameter is updated, it **MUST** call :code:`getParameterPtr()->incUpdate(callback);`. This is utilize for parameter update over multiple threads or multiple machines.
- Then it computes the gradients of the transformation matrices and inputs, and it calls :code:`incUpdate` for the corresponding parameter. This gives the framework the chance to know whether it has gathered all the gradient to one parameter so that it can do some overlapping work (e.g., network communication)
Z
zhangjinchao01 已提交
200 201


202
.. code-block:: c++
Z
zhangjinchao01 已提交
203

W
wangjiang03 已提交
204 205
    void FullyConnectedLayer::backward(const UpdateCallback& callback) {
      /* Do derivation for activations.*/ {
206
        backwardActivation();
W
wangjiang03 已提交
207
      }
Z
zhangjinchao01 已提交
208

W
wangjiang03 已提交
209
      if (biases_ && biases_->getWGrad()) {
210
        biases_->getWGrad()->collectBias(*getOutputGrad(), 1);
Z
zhangjinchao01 已提交
211

212
        biases_->getParameterPtr()->incUpdate(callback);
W
wangjiang03 已提交
213
      }
Z
zhangjinchao01 已提交
214

W
wangjiang03 已提交
215 216 217
      bool syncFlag = hl_get_sync_flag();

      for (size_t i = 0; i != inputLayers_.size(); ++i) {
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
        /* Calculate the W-gradient for the current layer */
        if (weights_[i]->getWGrad()) {
          MatrixPtr input_T = getInputValue(i)->getTranspose();
          MatrixPtr oGrad = getOutputGrad();
          {
            weights_[i]->getWGrad()->mul(input_T, oGrad, 1, 1);
          }
        }


        /* Calculate the input layers error */
        MatrixPtr preGrad = getInputGrad(i);
        if (NULL != preGrad) {
          MatrixPtr weights_T = weights_[i]->getW()->getTranspose();
          preGrad->mul(getOutputGrad(), weights_T, 1, 1);
        }

        {
          weights_[i]->getParameterPtr()->incUpdate(callback);
        }
Z
zhangjinchao01 已提交
238 239 240
      }
    }

241
The :code:`prefetch` function specifies the rows that need to be fetched from parameter server during training. It is only useful for remote sparse training. In remote sparse training, the full parameter matrix is stored distributedly at the parameter server. When the layer uses a batch for training, only a subset of locations of the input is non-zero in this batch. Thus, this layer only needs the rows of the transformation matrix corresponding to the locations of these non-zero entries. The :code:`prefetch` function specifies the ids of these rows.
Z
zhangjinchao01 已提交
242

243
Most of the layers do not need remote sparse training function. You do not need to override this function in this case.
W
wangjiang03 已提交
244

245
.. code-block:: c++
Z
zhangjinchao01 已提交
246

W
wangjiang03 已提交
247 248
    void FullyConnectedLayer::prefetch() {
      for (size_t i = 0; i != inputLayers_.size(); ++i) {
249 250 251 252 253 254
        auto* sparseParam =
            dynamic_cast<SparsePrefetchRowCpuMatrix*>(weights_[i]->getW().get());
        if (sparseParam) {
          MatrixPtr input = getInputValue(i);
          sparseParam->addRows(input);
        }
W
wangjiang03 已提交
255
      }
Z
zhangjinchao01 已提交
256 257
    }

258
Finally, you can use :code:`REGISTER_LAYER(fc, FullyConnectedLayer);` to register the layer. :code:`fc` is the identifier of the layer, and :code:`FullyConnectedLayer` is the class name of the layer.
Z
zhangjinchao01 已提交
259

260
.. code-block:: c++
Z
zhangjinchao01 已提交
261

W
wangjiang03 已提交
262 263
    namespace paddle {
    REGISTER_LAYER(fc, FullyConnectedLayer);
Z
zhangjinchao01 已提交
264 265
    }

X
Xin Pan 已提交
266
If the :code:`cpp` file is put into :code:`paddle/legacy/gserver/layers`, it will be automatically added to the compilation list.
Z
zhangjinchao01 已提交
267

268

W
wangjiang03 已提交
269
Write Gradient Check Unit Test
270
===============================
271

272
An easy way to verify the correctness of new layer's implementation is to write a gradient check unit test. Gradient check unit test utilizes finite difference method to verify the gradient of a layer. It modifies the input with a small perturbation :math:`\Delta x` and observes the changes of output :math:`\Delta y`, the gradient can be computed as :math:`\frac{\Delta y}{\Delta x }`. This gradient can be compared with the gradient computed by the :code:`backward` function of the layer to ensure the correctness of the gradient computation. Notice that the gradient check only tests the correctness of the gradient computation, it does not necessarily guarantee the correctness of the implementation of the :code:`forward` and :code:`backward` function. You need to write more sophisticated unit tests to make sure your layer is implemented correctly.
273

X
Xin Pan 已提交
274
All the gradient check unit tests are located in :code:`paddle/legacy/gserver/tests/test_LayerGrad.cpp`. You are recommended to put your test into a new test file if you are planning to write a new layer. The gradient test of the gradient check unit test of the fully connected layer is listed below. It has the following steps.
275 276

+ Create layer configuration. A layer configuration can include the following attributes:
277 278 279 280 281
   - size of the bias parameter. (4096 in our example)
   - type of the layer. (fc in our example)
   - size of the layer. (4096 in our example)
   - activation type. (softmax in our example)
   - dropout rate. (0.1 in our example)
282
+ configure the input of the layer. In our example, we have only one input.
283 284 285 286 287 288 289 290 291 292
   - type of the input (:code:`INPUT_DATA`) in our example. It can be one of the following types
       - :code:`INPUT_DATA`: dense vector.
       - :code:`INPUT_LABEL`: integer.
       - :code:`INPUT_DATA_TARGET`: dense vector, but it does not used to compute gradient.
       - :code:`INPUT_SEQUENCE_DATA`: dense vector with sequence information.
       - :code:`INPUT_HASSUB_SEQUENCE_DATA`: dense vector with both sequence and sub-sequence information.
       - :code:`INPUT_SEQUENCE_LABEL`: integer with sequence information.
       - :code:`INPUT_SPARSE_NON_VALUE_DATA`: 0-1 sparse data.
       - :code:`INPUT_SPARSE_FLOAT_VALUE_DATA`: float sparse data.
   - name of the input. (:code:`layer_0` in our example)
293 294 295
   - size of the input. (8192 in our example)
   - number of non-zeros, only useful for sparse inputs.
   - format of sparse data, only useful for sparse inputs.
296 297 298
+ each inputs needs to call :code:`config.layerConfig.add_inputs();` once.
+ call :code:`testLayerGrad` to perform gradient checks. It has the following arguments.
   - layer and input configurations. (:code:`config` in our example)
299
   - type of the layer. (:code:`fc` in our example)
300
   - batch size of the gradient check. (100 in our example)
301 302
   - whether the input is transpose. Most layers need to set it to :code:`false`. (:code:`false` in our example)
   - whether to use weights. Some layers or activations perform normalization so that the sum of their output is a constant. For example, the sum of output of a softmax activation is one. In this case, we cannot correctly compute the gradients using regular gradient check techniques. A weighted sum of the output, which is not a constant, is utilized to compute the gradients. (:code:`true` in our example, because the activation of a fully connected layer can be softmax)
W
wangjiang03 已提交
303

304
.. code-block:: c++
W
wangjiang03 已提交
305 306 307 308 309 310 311

    void testFcLayer(string format, size_t nnz) {
      // Create layer configuration.
      TestConfig config;
      config.biasSize = 4096;
      config.layerConfig.set_type("fc");
      config.layerConfig.set_size(4096);
312
      config.layerConfig.set_active_type("softmax");
W
wangjiang03 已提交
313 314 315
      config.layerConfig.set_drop_rate(0.1);
      // Setup inputs.
      config.inputDefs.push_back(
316 317
          {INPUT_DATA, "layer_0", 8192, nnz, ParaSparse(format)});
        config.layerConfig.add_inputs();
W
wangjiang03 已提交
318
      LOG(INFO) << config.inputDefs[0].sparse.sparse << " "
319
                << config.inputDefs[0].sparse.format;
W
wangjiang03 已提交
320
      for (auto useGpu : {false, true}) {
321 322
        testLayerGrad(config, "fc", 100, /* trans */ false, useGpu,
                      /* weight */ true);
W
wangjiang03 已提交
323 324
      }
    }
325

X
Xin Pan 已提交
326
If you are creating a new file for the test, such as :code:`paddle/legacy/gserver/tests/testFCGrad.cpp`, you need to add the file to :code:`paddle/legacy/gserver/tests/CMakeLists.txt`. An example is given below. All the unit tests will run when you execute the command :code:`make tests`. Notice that some layers might need high accuracy for the gradient check unit tests to work well. You need to configure :code:`WITH_DOUBLE` to `ON` when configuring cmake.
W
wangjiang03 已提交
327

328
.. code-block:: bash
329

W
wangjiang03 已提交
330
    add_unittest_without_exec(test_FCGrad
331 332 333
        test_FCGrad.cpp
        LayerGradUtil.cpp
        TestUtil.cpp)
Z
zhangjinchao01 已提交
334

W
wangjiang03 已提交
335
    add_test(NAME test_FCGrad
336
        COMMAND test_FCGrad)
W
wangjiang03 已提交
337 338 339


Implement Python Wrapper
340
========================
W
wangjiang03 已提交
341

342
Implementing Python wrapper allows us to use the added layer in configuration files. All the Python wrappers are in file :code:`python/paddle/trainer/config_parser.py`. An example of the Python wrapper for fully connected layer is listed below. It has the following steps:
W
wangjiang03 已提交
343

344 345 346
- Use :code:`@config_layer('fc')` at the decorator for all the Python wrapper class. :code:`fc` is the identifier of the layer.
- Implements :code:`__init__` constructor function.
	- It first call :code:`super(FCLayer, self).__init__(name, 'fc', size, inputs=inputs, **xargs)` base constructor function. :code:`FCLayer` is the Python wrapper class name, and :code:`fc` is the layer identifier name. They must be correct in order for the wrapper to work.
Z
zhangjinchao01 已提交
347
	- Then it computes the size and format (whether sparse) of each transformation matrix as well as the size.
348 349

.. code-block:: python
W
wangjiang03 已提交
350 351 352

    @config_layer('fc')
    class FCLayer(LayerBase):
353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370
        def __init__(
                self,
                name,
                size,
                inputs,
                bias=True,
                **xargs):
            super(FCLayer, self).__init__(name, 'fc', size, inputs=inputs, **xargs)
            for input_index in xrange(len(self.inputs)):
                input_layer = self.get_input_layer(input_index)
                psize = self.config.size * input_layer.size
                dims = [input_layer.size, self.config.size]
                format = self.inputs[input_index].format
                sparse = format == "csr" or format == "csc"
                if sparse:
                    psize = self.inputs[input_index].nnz
                self.create_input_parameter(input_index, psize, dims, sparse, format)
            self.create_bias_parameter(bias, self.config.size)
Z
zhangjinchao01 已提交
371 372 373

In network configuration, the layer can be specifies using the following code snippets. The arguments of this class are:

374 375 376 377 378 379 380
- :code:`name` is the name identifier of the layer instance.
- :code:`type` is the type of the layer, specified using layer identifier.
- :code:`size` is the output size of the layer.
- :code:`bias` specifies whether this layer instance has bias.
- :code:`inputs` specifies a list of layer instance names as inputs.

.. code-block:: python
W
wangjiang03 已提交
381 382

    Layer(
383 384 385 386 387
        name = "fc1",
        type = "fc",
        size = 64,
        bias = True,
        inputs = [Input("pool3")]
W
wangjiang03 已提交
388 389
    )

390
You are also recommended to implement a helper for the Python wrapper, which makes it easier to write models. You can refer to :code:`python/paddle/trainer_config_helpers/layers.py` for examples.