From 1ceffac51b1132a929c29af3fd5821f54d0fb2fe Mon Sep 17 00:00:00 2001 From: ceci3 <592712189@qq.com> Date: Fri, 13 Dec 2019 03:25:14 +0000 Subject: [PATCH] add search space doc --- demo/nas/README.md | 70 ++++++++++++++++++++++++++ demo/nas/search_space_doc.md | 98 ++++++++++++++++++++++++++++++++++++ 2 files changed, 168 insertions(+) create mode 100644 demo/nas/README.md create mode 100644 demo/nas/search_space_doc.md diff --git a/demo/nas/README.md b/demo/nas/README.md new file mode 100644 index 00000000..f28dbd71 --- /dev/null +++ b/demo/nas/README.md @@ -0,0 +1,70 @@ +# 网络结构搜索示例 + +本示例介绍如何使用网络结构搜索接口，搜索到一个更小或者精度更高的模型，该文档仅介绍paddleslim中SANAS的使用及如何利用SANAS得到模型结构，完整示例代码请参考sa_nas_mobilenetv2.py或者block_sa_nas_mobilenetv2.py。 + +## 接口介绍 +请参考。 + +### 1. 配置搜索空间 +详细的搜索空间配置可以参考神经网络搜索API文档。 +``` +config = [('MobileNetV2Space')] + +``` + +### 2. 利用搜索空间初始化SANAS实例 +``` +from paddleslim.nas import SANAS + +sa_nas = SANAS( + config, + server_addr=("", 8881), + init_temperature=10.24, + reduce_rate=0.85, + search_steps=300, + is_server=True) + +``` + +### 3. 根据实例化的NAS得到当前的网络结构 +``` +archs = sa_nas.next_archs() +``` + +### 4. 根据得到的网络结构和输入构造训练和测试program +``` +import paddle.fluid as fluid + +train_program = fluid.Program() +test_program = fluid.Program() +startup_program = fluid.Program() + +with fluid.program_guard(train_program, startup_program): + data = fluid.data(name='data', shape=[None, 3, 32, 32], dtype='float32') + label = fluid.data(name='label', shape=[None, 1], dtype='int64') + for arch in archs: + data = arch(data) + output = fluid.layers.fc(data, 10) + softmax_out = fluid.layers.softmax(input=output, use_cudnn=False) + cost = fluid.layers.cross_entropy(input=softmax_out, label=label) + avg_cost = fluid.layers.mean(cost) + acc_top1 = fluid.layers.accuracy(input=softmax_out, label=label, k=1) + + test_program = train_program.clone(for_test=True) + sgd = fluid.optimizer.SGD(learning_rate=1e-3) + sgd.minimize(avg_cost) + +``` + +### 5. 根据构造的训练program添加限制条件 +``` +from paddleslim.analysis import flops + +if flops(train_program) > 321208544: + continue +``` + +### 6. 回传score +``` +sa_nas.reward(score) +``` diff --git a/demo/nas/search_space_doc.md b/demo/nas/search_space_doc.md new file mode 100644 index 00000000..6877ae70 --- /dev/null +++ b/demo/nas/search_space_doc.md @@ -0,0 +1,98 @@ +# paddleslim.nas 提供的搜索空间： + +## 1. 根据原本模型结构构造搜索空间： + +### 1.1 MobileNetV2Space + +### 1.2 MobileNetV1Space + +### 1.3 ResNetSpace + + +## 2. 根据相应模型的block构造搜索空间 + +### 2.1 MobileNetV1BlockSpace + +### 2.2 MobileNetV2BlockSpace + +### 2.3 ResNetBlockSpace + +### 2.4 InceptionABlockSpace + +### 2.5 InceptionCBlockSpace + + + + +# 自定义搜索空间(search space) + +自定义搜索空间类需要继承搜索空间基类并重写以下几部分： + 1. 初始化的tokens(`init_tokens`函数)，可以设置为自己想要的tokens列表, tokens列表中的每个数字指的是当前数字在相应的搜索列表中的索引。例如本示例中若tokens=[0, 3, 5]，则代表当前模型结构搜索到的通道数为[8, 40, 128]。 + 2. token中每个数字的搜索列表长度(`range_table`函数)，tokens中每个token的索引范围。 + 3. 根据token产生模型结构(`token2arch`函数)，根据搜索到的tokens列表产生模型结构。 + +以新增reset block为例说明如何构造自己的search space。自定义的search space不能和已有的search space同名。 + +```python +### 引入搜索空间基类函数和search space的注册类函数 +from .search_space_base import SearchSpaceBase +from .search_space_registry import SEARCHSPACE +import numpy as np + +### 需要调用注册函数把自定义搜索空间注册到space space中 +@SEARCHSPACE.register +### 定义一个继承SearchSpaceBase基类的搜索空间的类函数 +class ResNetBlockSpace2(SearchSpaceBase): + def __init__(self, input_size, output_size, block_num, block_mask): + ### 定义一些实际想要搜索的内容，例如：通道数、每个卷积的重复次数、卷积核大小等等 + ### self.filter_num 代表通道数的搜索列表 + self.filter_num = np.array([8, 16, 32, 40, 64, 128, 256, 512]) + + ### 定义初始化token，初始化token的长度根据传入的block_num或者block_mask的长度来得到的 + def init_tokens(self): + return [0] * 3 * len(self.block_mask) + + ### 定义 + def range_table(self): + return [len(self.filter_num)] * 3 * len(self.block_mask) + + def token2arch(self, tokens=None): + if tokens == None: + tokens = self.init_tokens() + + self.bottleneck_params_list = [] + for i in range(len(self.block_mask)): + self.bottleneck_params_list.append(self.filter_num[tokens[i * 3 + 0]], + self.filter_num[tokens[i * 3 + 1]], + self.filter_num[tokens[i * 3 + 2]], + 2 if self.block_mask[i] == 1 else 1) + + def net_arch(input): + for i, layer_setting in enumerate(self.bottleneck_params_list): + channel_num, stride = layer_setting[:-1], layer_setting[-1] + input = self._resnet_block(input, channel_num, stride, name='resnet_layer{}'.format(i+1)) + + return input + + return net_arch + + ### 构造具体block的操作 + def _resnet_block(self, input, channel_num, stride, name=None): + shortcut_conv = self._shortcut(input, channel_num[2], stride, name=name) + input = self._conv_bn_layer(input=input, num_filters=channel_num[0], filter_size=1, act='relu', name=name + '_conv0') + input = self._conv_bn_layer(input=input, num_filters=channel_num[1], filter_size=3, stride=stride, act='relu', name=name + '_conv1') + input = self._conv_bn_layer(input=input, num_filters=channel_num[2], filter_size=1, name=name + '_conv2') + return fluid.layers.elementwise_add(x=shortcut_conv, y=input, axis=0, name=name+'_elementwise_add') + + def _shortcut(self, input, channel_num, stride, name=None): + channel_in = input.shape[1] + if channel_in != channel_num or stride != 1: + return self.conv_bn_layer(input, num_filters=channel_num, filter_size=1, stride=stride, name=name+'_shortcut') + else: + return input + + def _conv_bn_layer(self, input, num_filters, filter_size, stride=1, padding='SAME', act=None, name=None): + conv = fluid.layers.conv2d(input, num_filters, filter_size, stride, name=name+'_conv') + bn = fluid.layers.batch_norm(conv, act=act, name=name+'_bn') + return bn +``` -- GitLab