fix for python3.7 (#45)

* update fix

* fix init token

* fix bug

* update

* update doc

demo/nas/README.md

@@ -2,69 +2,25 @@
 
 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<a href='../../../paddleslim/nas/nas_api.md'>神经网络搜索API文档</a>。
-```
-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)
-
-```
+## 数据准备
+本示例默认使用cifar10数据，cifar10数据会根据调用的paddle接口自动下载，无需额外准备。
 
-### 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()
+## 接口介绍
+请参考<a href='../../docs/docs/api/nas_api.md'>神经网络搜索API文档</a>。
 
-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)
+本示例为在MobileNetV2的搜索空间上搜索FLOPs更小的模型。
+## 1 搜索空间配置
+默认搜索空间为`MobileNetV2`，详细的搜索空间配置请参考<a href='../../docs/docs/search_space.md'>搜索空间配置文档</a>。
 
-    test_program = train_program.clone(for_test=True)
-    sgd = fluid.optimizer.SGD(learning_rate=1e-3)
-    sgd.minimize(avg_cost)
-    
-```
+## 2 启动训练
 
-### 5. 根据构造的训练program添加限制条件
+### 2.1 启动基于MobileNetV2初始模型结构构造搜索空间的实验
+```shell
+CUDA_VISIBLE_DEVICES=0 python sa_nas_mobilenetv2.py
 ```
-from paddleslim.analysis import flops
 
-if flops(train_program) > 321208544:
-    continue
-```
 
-### 6. 回传score
-```
-sa_nas.reward(score)
+### 2.2 启动基于MobileNetV2的block构造搜索空间的实验
+```shell
+CUDA_VISIBLE_DEVICES=0 python block_sa_nas_mobilenetv2.py
 ```

docs/docs/api/nas_api.md

@@ -32,7 +32,7 @@ paddleslim.nas.SANAS(configs, server_addr=("", 8881), init_temperature=None, red
 ```python
 from paddleslim.nas import SANAS
 config = [('MobileNetV2Space')]
-sanas = SANAS(config=config)
+sanas = SANAS(configs=config)
 ```
 
 !!! note "Note"
@@ -48,26 +48,6 @@ sanas = SANAS(config=config)
     - 初始化token如果是随机生成的话，代表初始化token是一个比较差的token，SA算法可以处于一种不稳定的阶段进行搜索，尽可能的随机探索所有可能得token，从而找到一个较好的token。初始温度可以设置的高一些，例如设置为1000，退火率相对设置的小一些。
 
 
-paddlesim.nas.SANAS.tokens2arch(tokens)
-: 通过一组token得到实际的模型结构，一般用来把搜索到最优的token转换为模型结构用来做最后的训练。tokens的形式是一个列表，tokens映射到搜索空间转换成相应的网络结构，一组token对应唯一的一个网络结构。
-
-**参数：**
-
-- **tokens(list):** - 一组token。
-
-**返回：**
-根据传入的token得到一个模型结构实例。
-
-**示例代码：**
-```python
-import paddle.fluid as fluid
-input = fluid.data(name='input', shape=[None, 3, 32, 32], dtype='float32')
-archs = sanas.token2arch(tokens)
-for arch in archs:
-    output = arch(input)
-    input = output
-```
-
 paddleslim.nas.SANAS.next_archs()
 : 获取下一组模型结构。
 
@@ -84,7 +64,6 @@ for arch in archs:
     input = output
 ```
 
-
 paddleslim.nas.SANAS.reward(score)
 : 把当前模型结构的得分情况回传。
 
@@ -95,6 +74,27 @@ paddleslim.nas.SANAS.reward(score)
 **返回：**
 模型结构更新成功或者失败，成功则返回`True`，失败则返回`False`。
 
+
+paddlesim.nas.SANAS.tokens2arch(tokens)
+: 通过一组token得到实际的模型结构，一般用来把搜索到最优的token转换为模型结构用来做最后的训练。tokens的形式是一个列表，tokens映射到搜索空间转换成相应的网络结构，一组token对应唯一的一个网络结构。
+
+**参数：**
+
+- **tokens(list):** - 一组token。
+
+**返回：**
+根据传入的token得到一个模型结构实例。
+
+**示例代码：**
+```python
+import paddle.fluid as fluid
+input = fluid.data(name='input', shape=[None, 3, 32, 32], dtype='float32')
+archs = sanas.token2arch(tokens)
+for arch in archs:
+    output = arch(input)
+    input = output
+```
+
 paddleslim.nas.SANAS.current_info()
 : 返回当前token和搜索过程中最好的token和reward。
 

paddleslim/common/controller_client.py

@@ -71,3 +71,13 @@ class ControllerClient(object):
         tokens = socket_client.recv(1024).decode()
         tokens = [int(token) for token in tokens.strip("\n").split(",")]
         return tokens
+
+    def request_current_info(self):
+        """
+        Request for current information.
+        """
+        socket_client = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+        socket_client.connect((self.server_ip, self.server_port))
+        socket_client.send("current_info".encode())
+        current_info = socket_client.recv(1024).decode()
+        return eval(current_info)

paddleslim/common/controller_server.py

@@ -90,10 +90,18 @@ class ControllerServer(object):
                     (self._search_steps))) and not self._closed:
                 conn, addr = self._socket_server.accept()
                 message = conn.recv(1024).decode()
+                _logger.debug(message)
                 if message.strip("\n") == "next_tokens":
                     tokens = self._controller.next_tokens()
                     tokens = ",".join([str(token) for token in tokens])
                     conn.send(tokens.encode())
+                elif message.strip("\n") == "current_info":
+                    current_info = dict()
+                    current_info['best_tokens'] = self._controller.best_tokens
+                    current_info['best_reward'] = self._controller.max_reward
+                    current_info[
+                        'current_tokens'] = self._controller.current_tokens
+                    conn.send(str(current_info).encode())
                 else:
                     _logger.debug("recv message from {}: [{}]".format(addr,
                                                                       message))

paddleslim/common/sa_controller.py

@@ -81,7 +81,7 @@ class SAController(EvolutionaryController):
         self._iter = iters
         self._checkpoints = checkpoints
         self._searched = searched if searched != None else dict()
-        self._current_token = init_tokens
+        self._current_tokens = init_tokens
 
     def __getstate__(self):
         d = {}

paddleslim/nas/sa_nas.py

@@ -162,10 +162,7 @@ class SANAS(object):
         Returns:
             dict<name, value>: a dictionary include best tokens, best reward and current reward.
         """
-        current_dict = dict()
-        current_dict['best_tokens'] = self._controller.best_tokens
-        current_dict['best_reward'] = self._controller.max_reward
-        current_dict['current_tokens'] = self._controller.current_tokens
+        current_dict = self._controller_client.request_current_info()
         return current_dict
 
     def next_archs(self):

paddleslim/nas/search_space/base_layer.py

@@ -19,7 +19,7 @@ from paddle.fluid.param_attr import ParamAttr
 def conv_bn_layer(input,
                   filter_size,
                   num_filters,
-                  stride,
+                  stride=1,
                   padding='SAME',
                   num_groups=1,
                   act=None,
@@ -52,9 +52,9 @@ def conv_bn_layer(input,
         bias_attr=False)
     bn_name = name + '_bn'
     return fluid.layers.batch_norm(
-               input=conv,
-               act = act,
-               param_attr=ParamAttr(name=bn_name + '_scale'),
-               bias_attr=ParamAttr(name=bn_name + '_offset'),
-               moving_mean_name=bn_name + '_mean',
-               moving_variance_name=bn_name + '_variance')
+        input=conv,
+        act=act,
+        param_attr=ParamAttr(name=bn_name + '_scale'),
+        bias_attr=ParamAttr(name=bn_name + '_offset'),
+        moving_mean_name=bn_name + '_mean',
+        moving_variance_name=bn_name + '_variance')

paddleslim/nas/search_space/inception_block.py

@@ -58,7 +58,7 @@ class InceptionABlockSpace(SearchSpaceBase):
         """
         The initial token.
         """
-        return get_random_tokens(self.range_table)
+        return get_random_tokens(self.range_table())
 
     def range_table(self):
         """
@@ -175,7 +175,7 @@ class InceptionABlockSpace(SearchSpaceBase):
                 input = self._inceptionA(
                     input,
                     A_tokens=filter_nums,
-                    filter_size=filter_size,
+                    filter_size=int(filter_size),
                     stride=stride,
                     pool_type=pool_type,
                     name='inceptionA_{}'.format(i + 1))
@@ -287,7 +287,7 @@ class InceptionCBlockSpace(SearchSpaceBase):
         """
         The initial token.
         """
-        return get_random_tokens(self.range_table)
+        return get_random_tokens(self.range_table())
 
     def range_table(self):
         """
@@ -408,13 +408,13 @@ class InceptionCBlockSpace(SearchSpaceBase):
                 pool_type = 'avg' if layer_setting[11] == 0 else 'max'
                 if stride == 2:
                     layer_count += 1
-                if check_points((layer_count - 1) in return_block):
+                if check_points((layer_count - 1), return_block):
                     mid_layer[layer_count - 1] = input
 
                 input = self._inceptionC(
                     input,
                     C_tokens=filter_nums,
-                    filter_size=filter_size,
+                    filter_size=int(filter_size),
                     stride=stride,
                     pool_type=pool_type,
                     name='inceptionC_{}'.format(i + 1))

paddleslim/nas/search_space/mobilenet_block.py

@@ -60,7 +60,7 @@ class MobileNetV2BlockSpace(SearchSpaceBase):
         self.scale = scale
 
     def init_tokens(self):
-        return get_random_tokens(self.range_table)
+        return get_random_tokens(self.range_table())
 
     def range_table(self):
         range_table_base = []
@@ -153,7 +153,7 @@ class MobileNetV2BlockSpace(SearchSpaceBase):
                     c=int(c * self.scale),
                     n=n,
                     s=s,
-                    k=k,
+                    k=int(k),
                     name='mobilenetv2_' + str(i + 1))
                 in_c = int(c * self.scale)
 
@@ -289,9 +289,11 @@ class MobileNetV1BlockSpace(SearchSpaceBase):
                  scale=1.0):
         super(MobileNetV1BlockSpace, self).__init__(input_size, output_size,
                                                     block_num, block_mask)
-        # use input_size and output_size to compute self.downsample_num
-        self.downsample_num = compute_downsample_num(self.input_size,
-                                                     self.output_size)
+
+        if self.block_mask == None:
+            # use input_size and output_size to compute self.downsample_num
+            self.downsample_num = compute_downsample_num(self.input_size,
+                                                         self.output_size)
         if self.block_num != None:
             assert self.downsample_num <= self.block_num, 'downsample numeber must be LESS THAN OR EQUAL TO block_num, but NOW: downsample numeber is {}, block_num is {}'.format(
                 self.downsample_num, self.block_num)
@@ -305,7 +307,7 @@ class MobileNetV1BlockSpace(SearchSpaceBase):
         self.scale = scale
 
     def init_tokens(self):
-        return get_random_tokens(self.range_table)
+        return get_random_tokens(self.range_table())
 
     def range_table(self):
         range_table_base = []
@@ -383,7 +385,7 @@ class MobileNetV1BlockSpace(SearchSpaceBase):
                     num_filters2=filter_num2,
                     stride=stride,
                     scale=self.scale,
-                    kernel_size=kernel_size,
+                    kernel_size=int(kernel_size),
                     name='mobilenetv1_{}'.format(str(i + 1)))
 
             if return_mid_layer:

paddleslim/nas/search_space/mobilenetv1.py

@@ -191,7 +191,7 @@ class MobileNetV1Space(SearchSpaceBase):
                     num_groups=filter_num1,
                     stride=stride,
                     scale=self.scale,
-                    kernel_size=kernel_size,
+                    kernel_size=int(kernel_size),
                     name='mobilenetv1_{}'.format(str(i + 1)))
 
             ### return_block and end_points means block num

paddleslim/nas/search_space/mobilenetv2.py

@@ -182,7 +182,7 @@ class MobileNetV2Space(SearchSpaceBase):
                     c=int(c * self.scale),
                     n=n,
                     s=s,
-                    k=k,
+                    k=int(k),
                     name='mobilenetv2_conv' + str(i))
                 in_c = int(c * self.scale)
 

paddleslim/nas/search_space/resnet_block.py

@@ -32,9 +32,10 @@ class ResNetBlockSpace(SearchSpaceBase):
     def __init__(self, input_size, output_size, block_num, block_mask=None):
         super(ResNetBlockSpace, self).__init__(input_size, output_size,
                                                block_num, block_mask)
-        # use input_size and output_size to compute self.downsample_num
-        self.downsample_num = compute_downsample_num(self.input_size,
-                                                     self.output_size)
+        if self.block_mask == None:
+            # use input_size and output_size to compute self.downsample_num
+            self.downsample_num = compute_downsample_num(self.input_size,
+                                                         self.output_size)
         if self.block_num != None:
             assert self.downsample_num <= self.block_num, 'downsample numeber must be LESS THAN OR EQUAL TO block_num, but NOW: downsample numeber is {}, block_num is {}'.format(
                 self.downsample_num, self.block_num)
@@ -44,7 +45,7 @@ class ResNetBlockSpace(SearchSpaceBase):
         self.k_size = np.array([3, 5])
 
     def init_tokens(self):
-        return get_random_tokens(self.range_table)
+        return get_random_tokens(self.range_table())
 
     def range_table(self):
         range_table_base = []
@@ -133,7 +134,7 @@ class ResNetBlockSpace(SearchSpaceBase):
                     num_filters1=filter_num1,
                     num_filters2=filter_num3,
                     num_filters3=filter_num3,
-                    kernel_size=k_size,
+                    kernel_size=int(k_size),
                     repeat1=repeat1,
                     repeat2=repeat2,
                     stride=stride,