add maxout layer, including interface and unittest (#229)
* add maxout layer, including interface and unittest * follow maxout comments * auto setting channels * fix unittest bug in test_RecurrentGradientMachineavx_docs
luotao1
8 years ago
committed by
qingqing01
parent
e83950b0d2
commit
3dd8c9bea4
@ -0,0 +1,87 @@
|
||||
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
|
||||
|
||||
Licensed under the Apache License, Version 2.0 (the "License");
|
||||
you may not use this file except in compliance with the License.
|
||||
You may obtain a copy of the License at
|
||||
|
||||
http://www.apache.org/licenses/LICENSE-2.0
|
||||
|
||||
Unless required by applicable law or agreed to in writing, software
|
||||
distributed under the License is distributed on an "AS IS" BASIS,
|
||||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
See the License for the specific language governing permissions and
|
||||
limitations under the License. */
|
||||
|
||||
#include "MaxOutLayer.h"
|
||||
#include "hl_gpu.h"
|
||||
#include "hl_cnn.h"
|
||||
|
||||
namespace paddle {
|
||||
|
||||
REGISTER_LAYER(maxout, MaxOutLayer);
|
||||
|
||||
size_t MaxOutLayer::getSize() {
|
||||
const MaxOutConfig& maxoutConf = config_.inputs(0).maxout_conf();
|
||||
imgSizeH_ = inputLayers_[0]->getOutput().getFrameHeight();
|
||||
imgSizeW_ = inputLayers_[0]->getOutput().getFrameWidth();
|
||||
if (imgSizeH_ == 0) {
|
||||
imgSizeH_ = maxoutConf.img_size_y();
|
||||
}
|
||||
if (imgSizeW_ == 0) {
|
||||
imgSizeW_ = maxoutConf.img_size_x();
|
||||
}
|
||||
|
||||
featLen_ = imgSizeH_ * imgSizeW_;
|
||||
size_t layerSize = featLen_ * outputChannels_;
|
||||
|
||||
getOutput().setFrameHeight(imgSizeH_);
|
||||
getOutput().setFrameWidth(imgSizeW_);
|
||||
|
||||
return layerSize;
|
||||
}
|
||||
|
||||
bool MaxOutLayer::init(const LayerMap& layerMap,
|
||||
const ParameterMap& parameterMap) {
|
||||
/* Initialize the basic parent class */
|
||||
Layer::init(layerMap, parameterMap);
|
||||
|
||||
/* the size of inputs for maxout-layer is 1 */
|
||||
CHECK_EQ(config_.inputs_size(), 1UL);
|
||||
|
||||
const MaxOutConfig& conf = config_.inputs(0).maxout_conf();
|
||||
groups_ = conf.groups();
|
||||
channels_ = conf.channels();
|
||||
CHECK_EQ(channels_ % groups_, 0UL);
|
||||
outputChannels_ = channels_ / groups_;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
void MaxOutLayer::forward(PassType passType) {
|
||||
Layer::forward(passType);
|
||||
|
||||
/* malloc memory for the output_ if necessary */
|
||||
/* note: one sample correspond to one column */
|
||||
size_t batchSize = getInput(0).getBatchSize();
|
||||
size_t size = getSize();
|
||||
resetOutput(batchSize, size);
|
||||
MatrixPtr inputV = getInputValue(0);
|
||||
MatrixPtr outV = getOutputValue();
|
||||
|
||||
IVector::resizeOrCreate(maxoutId_, size * batchSize, useGpu_);
|
||||
outV->maxoutForward(*inputV, *maxoutId_, outputChannels_, groups_);
|
||||
}
|
||||
|
||||
void MaxOutLayer::backward(const UpdateCallback& callback) {
|
||||
(void)callback;
|
||||
|
||||
/* Do derivation */
|
||||
MatrixPtr inputG = getInputGrad(0);
|
||||
MatrixPtr outG = getOutputGrad();
|
||||
|
||||
if (inputG) {
|
||||
inputG->maxoutBackward(*outG, *maxoutId_, outputChannels_, groups_);
|
||||
}
|
||||
}
|
||||
|
||||
} // namespace paddle
|
||||
@ -0,0 +1,54 @@
|
||||
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
|
||||
|
||||
Licensed under the Apache License, Version 2.0 (the "License");
|
||||
you may not use this file except in compliance with the License.
|
||||
You may obtain a copy of the License at
|
||||
|
||||
http://www.apache.org/licenses/LICENSE-2.0
|
||||
|
||||
Unless required by applicable law or agreed to in writing, software
|
||||
distributed under the License is distributed on an "AS IS" BASIS,
|
||||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
See the License for the specific language governing permissions and
|
||||
limitations under the License. */
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "Layer.h"
|
||||
#include "paddle/math/Matrix.h"
|
||||
|
||||
namespace paddle {
|
||||
|
||||
/**
|
||||
* A layer to do max out on conv layer output.
|
||||
* Input: output of a conv layer.
|
||||
* Output: feature map size same as input. Channel is (input channel) / groups.
|
||||
* So the num of channels should be able to devided by groups.
|
||||
*
|
||||
* The config file api is maxout_layer.
|
||||
*/
|
||||
|
||||
class MaxOutLayer : public Layer {
|
||||
protected:
|
||||
size_t groups_;
|
||||
size_t imgSizeH_, imgSizeW_;
|
||||
/// outputChannels_ = channels_ / groups_
|
||||
size_t channels_, outputChannels_;
|
||||
/// feature length = imgSizeH_ * imgSizeW_
|
||||
size_t featLen_;
|
||||
IVectorPtr maxoutId_;
|
||||
|
||||
public:
|
||||
/// return imgSizeH_ * imgSizeW_ * outputChannels_;
|
||||
size_t getSize();
|
||||
|
||||
explicit MaxOutLayer(const LayerConfig& config) : Layer(config) {}
|
||||
virtual ~MaxOutLayer() {}
|
||||
|
||||
bool init(const LayerMap& layerMap, const ParameterMap& parameterMap);
|
||||
|
||||
void forward(PassType passType);
|
||||
void backward(const UpdateCallback& callback = nullptr);
|
||||
};
|
||||
|
||||
} // namespace paddle
|
||||
@ -0,0 +1,30 @@
|
||||
from paddle.trainer_config_helpers import *
|
||||
|
||||
settings(
|
||||
batch_size=1000,
|
||||
learning_rate=1e-5
|
||||
)
|
||||
|
||||
data = data_layer(name='data', size=2304)
|
||||
|
||||
conv = img_conv_layer(input=data,
|
||||
filter_size = 3,
|
||||
num_channels=1,
|
||||
num_filters=16,
|
||||
padding=1,
|
||||
act=LinearActivation(),
|
||||
bias_attr=True)
|
||||
|
||||
maxout = maxout_layer(input=conv,
|
||||
num_channels=16,
|
||||
groups=2)
|
||||
|
||||
pool = img_pool_layer(input=maxout,
|
||||
num_channels=8,
|
||||
pool_size=2,
|
||||
stride=2,
|
||||
pool_type=MaxPooling())
|
||||
|
||||
fc = fc_layer(input=pool, size=384, bias_attr=False)
|
||||
|
||||
outputs(fc)
|
||||
Loading…
Reference in new issue