fix bugs of dygraph/similarity_net (#4322)

* Update README.md (#4267)

* test=develop (#4269)

* 3d use new api (#4275)

* PointNet++ and PointRCNN use new API

* Update Readme of Dygraph BERT (#4277)

Fix some typos.

* Update run_classifier_multi_gpu.sh (#4279)

remove the CUDA_VISIBLE_DEVICES

* Update README.md (#4280)

* add similarity_net dygraph

* fix similarity_net dygraph

* fix bugs of dygraph/similarity_net
Co-authored-by: Npkpk <xiyzhouang@gmail.com>
Co-authored-by: NKaipeng Deng <dengkaipeng@baidu.com>

PaddleNLP/similarity_net/run_classifier.py

@@ -532,4 +532,4 @@ if __name__ == "__main__":
         infer(conf_dict, args)
     else:
         raise ValueError(
-            "one of do_train and do_test and do_infer must be True")
+            "one of do_train and do_test and do_infer must be True")
\ No newline at end of file

dygraph/similarity_net/README.md

 # 短文本语义匹配
 ## 简介
 ### 任务说明
-短文本语义匹配(SimilarityNet, SimNet)是一个计算短文本相似度的框架，可以根据用户输入的两个文本，计算出相似度得分。SimNet框架在百度各产品上广泛应用，主要包括BOW、CNN、RNN、MMDNN等核心网络结构形式，提供语义相似度计算训练和预测框架，适用于信息检索、新闻推荐、智能客服等多个应用场景，帮助企业解决语义匹配问题。可通过[AI开放平台-短文本相似度](https://ai.baidu.com/tech/nlp_basic/simnet)线上体验。
-
-同时推荐用户参考[ IPython Notebook demo](https://aistudio.baidu.com/aistudio/projectDetail/124373)
+短文本语义匹配(SimilarityNet, SimNet)是一个计算短文本相似度的框架，可以根据用户输入的两个文本，计算出相似度得分。SimNet框架在百度各产品上广泛应用，主要包括BOW、CNN、RNN、MMDNN等核心网络结构形式，提供语义相似度计算训练和预测框架，适用于信息检索、新闻推荐、智能客服等多个应用场景，帮助企业解决语义匹配问题。
 
 ### 效果说明
-基于百度海量搜索数据，我们训练了一个SimNet-BOW-Pairwise语义匹配模型，在一些真实的FAQ问答场景中，该模型效果比基于字面的相似度方法AUC提升5%以上，我们基于百度自建测试集（包含聊天、客服等数据集）和语义匹配数据集(LCQMC)进行评测，效果如下表所示。LCQMC数据集以Accuracy为评测指标，而pairwise模型的输出为相似度，因此我们采用0.958作为分类阈值，相比于基线模型中网络结构同等复杂的CBOW模型（准确率为0.737），我们模型的准确率为0.7532。
+基于百度海量搜索数据，我们训练了一个SimNet-BOW-Pairwise语义匹配模型，在一些真实的FAQ问答场景中，该模型效果比基于字面的相似度方法AUC提升5%以上，我们基于百度自建测试集（包含聊天、客服等数据集）和进行评测，效果如下表所示。
 
 
 | 模型       | 百度知道         | ECOM        |QQSIM | UNICOM |
 |:-----------:|:-------------:|:-------------:|:-------------:|:-------------:|
 |   | AUC  | AUC | AUC|正逆序比|
-|BOW_Pairwise|0.6767|0.7329|0.7650|1.5630|
+|BOW_Pairwise|0.6815|0.7331|0.7638|1.5566|
+
 #### 测试集说明
 | 数据集       | 来源         | 垂类        |
 |:-----------:|:-------------:|:-------------:|
@@ -22,17 +21,21 @@
 |UNICOM|联通客服|客服|
 ## 快速开始
 #### 版本依赖
-本项目依赖于 Paddlepaddle Fluid 1.6，请参考[安装指南](http://www.paddlepaddle.org/#quick-start)进行安装。
+
+本项目依赖于 Paddlepaddle Fluid 1.7，请参考[安装指南](http://www.paddlepaddle.org/#quick-start)进行安装。
+
 
 python版本依赖python 2.7
 #### 安装代码
 克隆工具集代码库到本地
 ```shell
 git clone https://github.com/PaddlePaddle/models.git
-cd models/PaddleNLP/similarity_net
+
+cd models/dygraph/similarity_net
 ```
 #### 数据准备
-下载经过预处理的数据，运行命令后，data目录下会存在训练集数据示例、集数据示例、测试集数据示例，以及对应词索引字典（term2id.dict）。
+下载经过预处理的数据，运行命令后，data目录下会存在训练集数据示例、测试集数据示例，以及对应词索引字典（term2id.dict）。
+
 ```shell
 sh download_data.sh
 ```
@@ -46,6 +49,7 @@ python download.py dataset
 sh download_pretrained_model.sh
 ```
 或者
+
 ```
 python download.py model
 ```
@@ -146,6 +150,7 @@ python tokenizer.py --test_data_dir ./test.txt.utf8 --batch_size 1 > test.txt.ut
 ├── utils.py：定义了其他常用的功能函数
 ├── Config: 定义多种模型的配置文件
 ├── download.py: 下载数据及预训练模型脚本
+├── nets: 基于动态图的网络结构
 ```
 
 ### 如何训练
@@ -178,7 +183,7 @@ python run_classifier.py \
 
 i. 定义自己的网络结构
 
-用户可以在```./models/```下定义自己的模型；
+用户可以在```./nets/```下定义自己的模型；
 
 ii. 更改模型配置
 

dygraph/similarity_net/download.py

@@ -67,6 +67,7 @@ def download(url, filename, md5sum):
     retry = 0
     retry_limit = 3
     chunk_size = 4096
+
     while not (os.path.exists(filename) and md5file(filename) == md5sum):
         if retry < retry_limit:
             retry += 1
@@ -115,8 +116,8 @@ def download_dataset(dir_path):
 def download_model(dir_path):
     MODELS = {}
     BASE_URL = "https://baidu-nlp.bj.bcebos.com/"
-    CNN_NAME = "simnet_bow-pairwise-1.0.0.tar.gz"
-    CNN_MD5 = "199a3f3af31558edcc71c3b54ea5e129"
+    CNN_NAME = "simnet_bow_pairwise_dygraph.tar.gz"
+    CNN_MD5 = "30012af0ca8cdf0c613d8f56884f0f48"
     MODELS[CNN_NAME] = CNN_MD5
 
     if not os.path.exists(dir_path):

dygraph/similarity_net/download_pretrained_model.sh

@@ -2,9 +2,10 @@
 model_files_path="./model_files"
 
 #get pretrained_bow_pairwise_model
-wget --no-check-certificate https://baidu-nlp.bj.bcebos.com/simnet_bow-pairwise-1.0.0.tar.gz
+
+wget --no-check-certificate https://baidu-nlp.bj.bcebos.com/simnet_bow_pairwise_dygraph.tar.gz
 if [ ! -d $model_files_path ]; then
 	mkdir $model_files_path
 fi
-tar xzf simnet_bow-pairwise-1.0.0.tar.gz -C $model_files_path
-rm simnet_bow-pairwise-1.0.0.tar.gz
\ No newline at end of file
+tar xzf simnet_bow_pairwise_dygraph.tar.gz -C $model_files_path
+rm simnet_bow_pairwise_dygraph.tar.gz

dygraph/similarity_net/evaluate/evaluate_ecom.sh

@@ -10,7 +10,8 @@ CKPT_PATH=./model_files
 TEST_RESULT_PATH=./evaluate/ecom_test_result
 TASK_MODE='pairwise'
 CONFIG_PATH=./config/bow_pairwise.json
-INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/
+INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/bow_pairwise
+
 cd ..
 
 python ./run_classifier.py \

dygraph/similarity_net/evaluate/evaluate_qqsim.sh

@@ -10,7 +10,8 @@ CKPT_PATH=./model_files
 TEST_RESULT_PATH=./evaluate/qqsim_test_result
 TASK_MODE='pairwise'
 CONFIG_PATH=./config/bow_pairwise.json
-INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/
+INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/bow_pairwise
+
 cd ..
 
 python ./run_classifier.py \

dygraph/similarity_net/evaluate/evaluate_unicom.sh

@@ -10,7 +10,8 @@ CKPT_PATH=./model_files
 INFER_RESULT_PATH=./evaluate/unicom_infer_result
 TASK_MODE='pairwise'
 CONFIG_PATH=./config/bow_pairwise.json
-INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/
+INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/bow_pairwise
+
 
 python unicom_split.py
 cd ..

dygraph/similarity_net/evaluate/evaluate_zhidao.sh

@@ -10,7 +10,8 @@ CKPT_PATH=./model_files
 TEST_RESULT_PATH=./evaluate/zhidao_test_result
 TASK_MODE='pairwise'
 CONFIG_PATH=./config/bow_pairwise.json
-INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/
+INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/bow_pairwise
+
 cd ..
 
 python ./run_classifier.py \

dygraph/similarity_net/nets/bow.py

@@ -18,9 +18,10 @@ bow class
 import paddle_layers as layers
 from paddle import fluid
 from paddle.fluid.dygraph.base import to_variable
-from paddle.fluid.dygraph import Layer, Embedding
+from paddle.fluid.dygraph import Layer, Embedding, Linear
 import paddle.fluid.param_attr as attr
 uniform_initializer = lambda x: fluid.initializer.UniformInitializer(low=-x, high=x)
+
 class BOW(Layer):
     """
     BOW
@@ -35,10 +36,11 @@ class BOW(Layer):
         self.task_mode = conf_dict["task_mode"]
         self.emb_dim = conf_dict["net"]["emb_dim"]
         self.bow_dim = conf_dict["net"]["bow_dim"]
-        self.seq_len = 5
+        self.seq_len = conf_dict["seq_len"]
         self.emb_layer = layers.EmbeddingLayer(self.dict_size, self.emb_dim, "emb").ops()
-        self.bow_layer = layers.FCLayer(self.bow_dim, None, "fc").ops()
+        self.bow_layer = Linear(self.bow_dim, self.bow_dim)
         self.softmax_layer = layers.FCLayer(2, "softmax", "cos_sim").ops()
+    
 
     def forward(self, left, right):
         """
@@ -46,7 +48,6 @@ class BOW(Layer):
         """
         
         # embedding layer
-
         left_emb = self.emb_layer(left)
         right_emb = self.emb_layer(right)
         left_emb = fluid.layers.reshape(

dygraph/similarity_net/nets/cnn.py

@@ -35,7 +35,7 @@ class CNN(Layer):
         self.filter_size = conf_dict["net"]["filter_size"]
         self.num_filters = conf_dict["net"]["num_filters"]
         self.hidden_dim = conf_dict["net"]["hidden_dim"]
-        self.seq_len = 5
+        self.seq_len = conf_dict["seq_len"]
         self.channels = 1
         
         # layers

dygraph/similarity_net/nets/gru.py

@@ -43,7 +43,7 @@ class GRU(Layer):
         self.fc_layer = layers.FCLayer(self.hidden_dim, None, "fc").ops()
         self.proj_layer = Linear(input_dim = self.hidden_dim, output_dim=self.gru_dim*3)
         self.softmax_layer = layers.FCLayer(2, "softmax", "cos_sim").ops()
-        self.seq_len=5
+        self.seq_len=conf_dict["seq_len"]
 
     def forward(self, left, right):
         """

dygraph/similarity_net/nets/lstm.py

@@ -38,7 +38,7 @@ class LSTM(Layer):
         self.fc_layer = layers.FCLayer(self.hidden_dim, None, "fc").ops()
         self.softmax_layer = layers.FCLayer(2, "softmax", "cos_sim").ops()
         self.proj_layer = Linear(input_dim = self.hidden_dim, output_dim=self.lstm_dim*4)
-        self.seq_len = 5
+        self.seq_len = conf_dict["seq_len"]
 
 
     def forward(self, left, right):

dygraph/similarity_net/nets/mm_dnn.py

@@ -42,10 +42,11 @@ class MMDNN(Layer):
         self.dpool_size1 = int(config['net']['dpool_size_left'])
         self.dpool_size2 = int(config['net']['dpool_size_right'])
         self.hidden_size = int(config['net']['hidden_size'])
-    
-        self.seq_len1 = 5
+        self.seq_len = int(conf_dict["seq_len"])
+        self.seq_len1 = self.seq_len
         #int(config['max_len_left'])
-        self.seq_len2 = 5 #int(config['max_len_right'])
+        self.seq_len2 = self.seq_len 
+        #int(config['max_len_right'])
         self.task_mode = config['task_mode']
         self.zero_pad = True
         self.scale = False
@@ -130,14 +131,14 @@ class MMDNN(Layer):
         right_seq_encoder = fluid.layers.concat([right_lstm, right_reverse], axis=1)
   
         pad_value = fluid.layers.assign(input=np.array([0]).astype("float32"))
- 
-        left_seq_encoder = fluid.layers.reshape(left_seq_encoder, shape=[left_seq_encoder.shape[0]/5,5,-1])
-        right_seq_encoder = fluid.layers.reshape(right_seq_encoder, shape=[right_seq_encoder.shape[0]/5,5,-1])
+        left_seq_encoder = fluid.layers.reshape(left_seq_encoder, shape=[int(left_seq_encoder.shape[0]/self.seq_len),self.seq_len,-1])
+        right_seq_encoder = fluid.layers.reshape(right_seq_encoder, shape=[int(right_seq_encoder.shape[0]/self.seq_len),self.seq_len,-1])
         cross = fluid.layers.matmul(
             left_seq_encoder, right_seq_encoder, transpose_y=True)
   
-        left_lens=to_variable(np.array([5]))
-        right_lens=to_variable(np.array([5]))
+        left_lens=to_variable(np.array([self.seq_len]))
+        right_lens=to_variable(np.array([self.seq_len]))
+
 
         if self.match_mask:
             mask1 = fluid.layers.sequence_mask(
@@ -157,7 +158,8 @@ class MMDNN(Layer):
         if mask is not None:
             cross_mask = fluid.layers.stack(x=[mask] * self.kernel_size, axis=0)
             cross_mask = fluid.layers.stack(x=[cross] * conv.shape[1], axis=1)
-            conv = cross_mask * conv + (1 - cross_mask) * (-2**5 + 1)
+            conv = cross_mask * conv + (1 - cross_mask) * (-2**self.seq_len + 1)
+
         pool = self.pool_layer(conv)
         conv_pool_relu = fluid.layers.relu(pool)
 

dygraph/similarity_net/nets/paddle_layers.py

@@ -20,7 +20,7 @@ import inspect
 import six
 import sys
 from functools import partial
-
+from functools import reduce
 import numpy as np
 import paddle
 import paddle.fluid as fluid
@@ -1047,37 +1047,3 @@ class BasicGRUUnit(Layer):
 
         return new_hidden
 
-
-######  DELETE
-
-# @contextlib.contextmanager
-# def eager_guard(is_eager):
-#     if is_eager:
-#         with fluid.dygraph.guard():
-#             yield
-#     else:
-#         yield
-
-# # print(flatten(np.random.rand(2,8,8)))
-# random_seed = 123
-# np.random.seed(random_seed)
-# # print np.random.rand(2, 8)
-# batch_size = 2
-# seq_len = 8
-# hidden_size = 8
-# vocab_size, embed_dim, num_layers, hidden_size = 100, 8, 2, 8
-# import torch
-
-
-# with eager_guard(False):
-#     fluid.default_main_program().random_seed = random_seed
-#     fluid.default_startup_program().random_seed = random_seed
-#     lstm_cell = BasicLSTMUnit(hidden_size=8, input_size=8)
-#     lstm = RNN(cell=lstm_cell, time_major=True)
-#     #print lstm(inputs=to_variable(np.random.rand(2, 8, 8).astype("float32")))[0].numpy()
-#     executor.run(fluid.default_startup_program())
-#     x = fluid.data(name="x", shape=[None, None, 8], dtype="float32")
-#     out, _ = lstm(x)
-#     out = executor.run(feed={"x": np.random.rand(2, 8, 8).astype("float32")}, fetch_list=[out.name])[0]
-#     print np.array(out)
-

dygraph/similarity_net/reader.py

@@ -28,6 +28,16 @@ class SimNetProcessor(object):
         self.valid_label = np.array([])
         self.test_label = np.array([])
 
+        self.seq_len = args.seq_len
+    
+    def padding_text(self, x):
+        if len(x) < self.seq_len:
+            x += [0]*(self.seq_len-len(x))
+        if len(x) > self.seq_len:
+            x = x[0:self.seq_len] 
+        return x
+
+
     def get_reader(self, mode, epoch=0):
         """
         Get Reader
@@ -60,6 +70,10 @@ class SimNetProcessor(object):
                             query = [0]
                         if len(title) == 0:
                             title = [0]
+
+                        query = self.padding_text(query)
+                        title = self.padding_text(title)
+
                         yield [query, title]
             elif mode == "test":
                 with io.open(self.args.test_data_dir, "r", encoding="utf8") as file:
@@ -83,8 +97,10 @@ class SimNetProcessor(object):
                             query = [0]
                         if len(title) == 0:
                             title = [0]
-                        # query = np.array([x.reshape(-1,1) for x in query]).astype('int64')
-                        # title = np.array([x.reshape(-1,1) for x in title]).astype('int64')
+
+                        query = self.padding_text(query)
+                        title = self.padding_text(title)
+
                         yield [query, title]
             else:
                 for idx in range(epoch):
@@ -115,7 +131,11 @@ class SimNetProcessor(object):
                                 pos_title = [0]
                             if len(neg_title) == 0:
                                 neg_title = [0]
-                            
+                                
+                            query = self.padding_text(query)
+                            pos_title = self.padding_text(pos_title)
+                            neg_title = self.padding_text(neg_title)
+
                             yield [query, pos_title, neg_title]
 
         def reader_with_pointwise():
@@ -145,6 +165,10 @@ class SimNetProcessor(object):
                             query = [0]
                         if len(title) == 0:
                             title = [0]
+                            
+                        query = self.padding_text(query)
+                        title = self.padding_text(title)
+
                         yield [query, title]
             elif mode == "test":
                 with io.open(self.args.test_data_dir, "r", encoding="utf8") as file:
@@ -168,6 +192,10 @@ class SimNetProcessor(object):
                             query = [0]
                         if len(title) == 0:
                             title = [0]
+
+                        query = self.padding_text(query)
+                        title = self.padding_text(title)
+
                         yield [query, title]
             else:
                 for idx in range(epoch):
@@ -194,6 +222,10 @@ class SimNetProcessor(object):
                                 query = [0]
                             if len(title) == 0:
                                 title = [0]
+
+                            query = self.padding_text(query)
+                            title = self.padding_text(title)
+
                             yield [query, title, label]
 
         if self.args.task_mode == "pairwise":
@@ -223,6 +255,10 @@ class SimNetProcessor(object):
                     query = [0]
                 if len(title) == 0:
                     title = [0]
+
+                query = self.padding_text(query)
+                title = self.padding_text(title)
+
                 yield [query, title]
 
     def get_infer_data(self):

dygraph/similarity_net/run.sh

@@ -14,7 +14,9 @@ TEST_RESULT_PATH=./test_result
 INFER_RESULT_PATH=./infer_result
 TASK_MODE='pairwise'
 CONFIG_PATH=./config/bow_pairwise.json
-INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/
+
+INIT_CHECKPOINT=./model_files/simnet_bow_pairwise_pretrained_model/bow_pairwise
+
 
 
 # run_train

dygraph/similarity_net/run_classifier.py

@@ -96,6 +96,8 @@ def train(conf_dict, args):
         vocab = utils.load_vocab(args.vocab_path)
         # get vocab size
         conf_dict['dict_size'] = len(vocab)
+        conf_dict['seq_len'] = args.seq_len
+
         # Load network structure dynamically
         net = utils.import_class("./nets",
                                 conf_dict["net"]["module_name"],
@@ -302,6 +304,7 @@ def test(conf_dict, args):
     else:
         place = fluid.CPUPlace()
     with fluid.dygraph.guard(place):
+
         vocab = utils.load_vocab(args.vocab_path)
         simnet_process = reader.SimNetProcessor(args, vocab)
         test_pyreader = fluid.io.PyReader(capacity=16, return_list=True, use_double_buffer=False)
@@ -310,7 +313,9 @@ def test(conf_dict, args):
                 paddle.batch(get_test_examples, batch_size=args.batch_size),
                 place)
      
-        conf_dict['dict_size'] = len(vocab)
+
+        conf_dict['dict_size'] = len(vocab)         
+        conf_dict['seq_len'] = args.seq_len
 
         net = utils.import_class("./nets",
                                 conf_dict["net"]["module_name"],
@@ -329,18 +334,22 @@ def test(conf_dict, args):
                     left_feat, pos_score = net(left, pos_right)
                     pred = pos_score
                     # pred_list += list(pred.numpy())
-                    pred_list += list(map(lambda item: float(item[0]), pred.numpy()[0]))
+
+                    pred_list += list(map(lambda item: float(item[0]), pred.numpy()))
                     predictions_file.write(u"\n".join(
-                            map(lambda item: str((item[0] + 1) / 2), pred.numpy()[0])) + "\n")
+                            map(lambda item: str((item[0] + 1) / 2), pred.numpy())) + "\n")
+
             else:
                 for left, right in test_pyreader():
                     left = fluid.layers.reshape(left, shape=[-1, 1])
                     right = fluid.layers.reshape(right, shape=[-1, 1])
                     left_feat, pred = net(left, right)
                     # pred_list += list(pred.numpy())
-                    pred_list += list(map(lambda item: float(item[0]), pred.numpy()[0]))
+
+                    pred_list += list(map(lambda item: float(item[0]), pred.numpy()))
                     predictions_file.write(u"\n".join(
-                            map(lambda item: str(np.argmax(item)), pred.numpy()[0])) + "\n")
+                            map(lambda item: str(np.argmax(item)), pred.numpy())) + "\n")
+
 
             if args.task_mode == "pairwise":
                 pred_list = np.array(pred_list).reshape((-1, 1))
@@ -377,45 +386,48 @@ def infer(conf_dict, args):
         place = fluid.CPUPlace()
    
 
-    vocab = utils.load_vocab(args.vocab_path)
-    simnet_process = reader.SimNetProcessor(args, vocab)
-    get_infer_examples = simnet_process.get_infer_reader
-    infer_pyreader = fluid.io.PyReader(capacity=16, return_list=True, use_double_buffer=False)
-    infer_pyreader.decorate_sample_list_generator(
-                paddle.batch(get_infer_examples, batch_size=args.batch_size),
-                place) 
-    
-    conf_dict['dict_size'] = len(vocab)
-
-    net = utils.import_class("./nets",
-                             conf_dict["net"]["module_name"],
-                             conf_dict["net"]["class_name"])(conf_dict)
-    model, _ = fluid.dygraph.load_dygraph(args.init_checkpoint)
-    net.set_dict(model)
-    pred_list = []
-    if args.task_mode == "pairwise":
-        for left, pos_right in infer_pyreader():
-            left = fluid.layers.reshape(left, shape=[-1, 1])
-            pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1])
-               
-            left_feat, pos_score = net(left, pos_right)
-            pred = pos_score
-            preds_list += list(
-                        map(lambda item: str((item[0] + 1) / 2), pred.numpy()[0]))
+    with fluid.dygraph.guard(place):
+        vocab = utils.load_vocab(args.vocab_path)
+        simnet_process = reader.SimNetProcessor(args, vocab)
+        get_infer_examples = simnet_process.get_infer_reader
+        infer_pyreader = fluid.io.PyReader(capacity=16, return_list=True, use_double_buffer=False)
+        infer_pyreader.decorate_sample_list_generator(
+                    paddle.batch(get_infer_examples, batch_size=args.batch_size),
+                    place) 
         
-    else:
-        for left, right in infer_pyreader():
-            left = fluid.layers.reshape(left, shape=[-1, 1])
-            pos_right = fluid.layers.reshape(right, shape=[-1, 1])
-            left_feat, pred = net(left, right)
-            preds_list += map(lambda item: str(np.argmax(item)), pred.numpy()[0])
-      
-   
-    with io.open(args.infer_result_path, "w", encoding="utf8") as infer_file:
-        for _data, _pred in zip(simnet_process.get_infer_data(), preds_list):
-            infer_file.write(_data + "\t" + _pred + "\n")
-    logging.info("infer result saved in %s" %
-                 os.path.join(os.getcwd(), args.infer_result_path))
+        conf_dict['dict_size'] = len(vocab) 
+        conf_dict['seq_len'] = args.seq_len
+
+        net = utils.import_class("./nets",
+                                conf_dict["net"]["module_name"],
+                                conf_dict["net"]["class_name"])(conf_dict)
+        model, _ = fluid.dygraph.load_dygraph(args.init_checkpoint)
+        net.set_dict(model)
+        
+        pred_list = []
+        if args.task_mode == "pairwise":
+            for left, pos_right in infer_pyreader():
+                left = fluid.layers.reshape(left, shape=[-1, 1])
+                pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1])
+                
+                left_feat, pos_score = net(left, pos_right)
+                pred = pos_score
+                pred_list += list(
+                            map(lambda item: str((item[0] + 1) / 2), pred.numpy()))
+            
+        else:
+            for left, right in infer_pyreader():
+                left = fluid.layers.reshape(left, shape=[-1, 1])
+                pos_right = fluid.layers.reshape(right, shape=[-1, 1])
+                left_feat, pred = net(left, right)
+                pred_list += map(lambda item: str(np.argmax(item)), pred.numpy())
+        
+    
+        with io.open(args.infer_result_path, "w", encoding="utf8") as infer_file:
+            for _data, _pred in zip(simnet_process.get_infer_data(), pred_list):
+                infer_file.write(_data + "\t" + _pred + "\n")
+        logging.info("infer result saved in %s" %
+                    os.path.join(os.getcwd(), args.infer_result_path))
 
 
 def get_cards():

dygraph/similarity_net/utils.py

@@ -175,6 +175,7 @@ class ArgConfig(object):
         model_g.add_arg("output_dir", str, None, "Directory path to save checkpoints")
         model_g.add_arg("task_mode", str, None, "task mode: pairwise or pointwise")
 
+
         train_g = ArgumentGroup(parser, "training", "training options.")
         train_g.add_arg("epoch", int, 10, "Number of epoches for training.")
         train_g.add_arg("save_steps", int, 200, "The steps interval to save checkpoints.")
@@ -193,6 +194,8 @@ class ArgConfig(object):
         data_g.add_arg("infer_data_dir", str, None, "Directory path to infer data.")
         data_g.add_arg("vocab_path", str, None, "Vocabulary path.")
         data_g.add_arg("batch_size", int, 32, "Total examples' number in batch for training.")
+        data_g.add_arg("seq_len", int, 32, "The length of each sentence.")
+
 
         run_type_g = ArgumentGroup(parser, "run_type", "running type options.")
         run_type_g.add_arg("use_cuda", bool, False, "If set, use GPU for training.")