readme修改

FarmingTong · Mar 6, 2022 · bfd3914 · bfd3914
1 parent d4bb196
commit bfd3914
Show file tree

Hide file tree

Showing 22 changed files with 33 additions and 19 deletions.
diff --git a/README.md b/README.md
@@ -66,8 +66,13 @@ bert_crf_token_classification_raw/
 2、在config.py中修改dir_name
 
 # bert 训练曲线(20epoch)
-![image](data/med_data/dev_f1.png)
-![image](data/med_data/loss.png)
+
+model   |  acc             |  loss
+:-------------------------:|:-------------------------:|:-------------------------:
+bert | ![](data/cner/bert_dev_f1.png)  |  ![](data/cner/bert_loss.png)
+roberta | ![](data/cner/roberta_dev_f1.png)  |  ![](data/cner/roberta_loss.png)
+ernie  | ![](data/cner/ernie_dev_f1.png)  |  ![](data/cner/ernie_loss.png)
+albert  | ![](data/cner/albert_dev_f1.png)  |  ![](data/cner/albert_loss.png)
 
 # 具体参数可看config.py
 ```

diff --git a/config.py b/config.py
@@ -15,18 +15,29 @@
              'transformers.BertConfig',
              'bert-base-chinese'  # 使用模型
              ),
-
+    'ernie': (
+        'transformers.AutoTokenizer',
+        'transformers.BertModel',
+        'transformers.AutoConfig',
+        "nghuyong/ernie-1.0",  # 使用模型参数
+    ),
     'roberta': (
         'transformers.BertTokenizer',
         'transformers.RobertaModel',
         'transformers.RobertaConfig',
         'hfl/chinese-roberta-wwm-ext',
     ),
-    'albert': ('transformers.BertTokenizer',
+    'albert': ('transformers.AutoTokenizer',
                'transformers.AlbertModel',
-               'transformers.AlbertConfig'
+               'transformers.AutoConfig',
+               "voidful/albert_chinese_tiny",  # 使用模型参数
                ),
 }
+MODEL = 'roberta'
+# MODEL = 'ernie'
+# MODEL = 'albert'
+# MODEL = 'bert'
+
 epochs = 20
 batch_size = 32
 # batch_size = 1
@@ -46,7 +57,7 @@
 # 切换任务时 数据配置
 csv_rows = ['raw_sen', 'label']  # csv的行标题，文本 和 类（目前类必须是列表）
 
-dir_name = 'med_data'
+dir_name = 'cner'
 train_file = f"data/{dir_name}/train.csv"
 dev_file = f"data/{dir_name}/dev.csv"
 # dev_file = f"data/{dir_name}/train.csv"
@@ -59,7 +70,6 @@
 test_pred_out = f"data/{dir_name}/test_data_predict.csv"
 # csv_encoding = 'gbk'
 
-MODEL = 'bert'
 
 PREFIX = ''
 # max_src_length = 400

diff --git a/data/cner/albert_dev_f1.png b/data/cner/albert_dev_f1.png
diff --git a/data/cner/albert_loss.png b/data/cner/albert_loss.png
diff --git a/data/med_data/dev_f1.png → data/cner/bert_dev_f1.png b/data/med_data/dev_f1.png → data/cner/bert_dev_f1.png
diff --git a/data/med_data/loss.png → data/cner/bert_loss.png b/data/med_data/loss.png → data/cner/bert_loss.png
diff --git a/data/med_data/dev.char.bmes → data/cner/dev.char.bmes b/data/med_data/dev.char.bmes → data/cner/dev.char.bmes
diff --git a/data/med_data/dev.csv → data/cner/dev.csv b/data/med_data/dev.csv → data/cner/dev.csv
diff --git a/data/cner/ernie_dev_f1.png b/data/cner/ernie_dev_f1.png
diff --git a/data/cner/ernie_loss.png b/data/cner/ernie_loss.png
diff --git a/data/med_data/label_2_id.json → data/cner/label_2_id.json b/data/med_data/label_2_id.json → data/cner/label_2_id.json
diff --git a/data/cner/roberta_dev_f1.png b/data/cner/roberta_dev_f1.png
diff --git a/data/cner/roberta_loss.png b/data/cner/roberta_loss.png
diff --git a/data/med_data/test.char.bmes → data/cner/test.char.bmes b/data/med_data/test.char.bmes → data/cner/test.char.bmes
diff --git a/data/med_data/test.csv → data/cner/test.csv b/data/med_data/test.csv → data/cner/test.csv
diff --git a/data/med_data/test_data_predict.csv → data/cner/test_data_predict.csv b/data/med_data/test_data_predict.csv → data/cner/test_data_predict.csv
diff --git a/data/med_data/train.char.bmes → data/cner/train.char.bmes b/data/med_data/train.char.bmes → data/cner/train.char.bmes
diff --git a/data/med_data/train.csv → data/cner/train.csv b/data/med_data/train.csv → data/cner/train.csv
diff --git a/data/txt2csv.py b/data/txt2csv.py
@@ -69,7 +69,7 @@ def change(file_name, file_out, dict_file, split_=' ', split_3=False):
 
 if __name__ == '__main__':
     json_dict = 'med_data/label_2_id.json'
-    get_dict('med_data/train.char.bmes', json_dict)  # 生成label2id字典，存储到json_dict
-    change('med_data/train.char.bmes', 'med_data/train.csv', json_dict)
-    change('med_data/test.char.bmes', 'med_data/test.csv', json_dict)
-    change('med_data/dev.char.bmes', 'med_data/dev.csv', json_dict)
+    get_dict('cner/train.char.bmes', json_dict)  # 生成label2id字典，存储到json_dict
+    change('cner/train.char.bmes', 'cner/train.csv', json_dict)
+    change('cner/test.char.bmes', 'cner/test.csv', json_dict)
+    change('cner/dev.char.bmes', 'cner/dev.csv', json_dict)
diff --git a/infer.py b/infer.py
@@ -15,7 +15,6 @@
 
 from torch.utils.data import Dataset, DataLoader
 import pandas as pd
-import torch
 from tqdm import tqdm
 from transformers import BertTokenizer
 
@@ -24,7 +23,7 @@
 
 label_2_id = load_json(json_dict)
 
-dir_name = 'med_data'
+dir_name = 'cner'
 target_file = f'models/{dir_name}/best.pth.tar'  # 模型存储路径
 label_file = f'data/{dir_name}/label2id.json'
 bert_path_or_name = 'bert-base-chinese'  # 使用模型

diff --git a/model.py b/model.py
@@ -1,8 +1,5 @@
 # -*- coding: utf-8 -*-
-import torch
 from torch import nn
-# from transformers import BertForSequenceClassification, BertConfig, RobertaForSequenceClassification, RobertaConfig, \
-#     AutoModelForSequenceClassification, AutoConfig
 from config import *
 from utils import eval_object
 
@@ -435,7 +432,10 @@ def __init__(self):
         self.device = torch.device("cuda")
         # self.dropout = nn.Dropout(hidden_dropout_prob)
         self.dropout = nn.Dropout(0.1)
-        self.classifier = nn.Linear(768, num_labels)
+        if MODEL == 'albert':
+            self.classifier = nn.Linear(312, num_labels)
+        else:
+            self.classifier = nn.Linear(768, num_labels)
         self.crf = CRF(num_tags=num_labels, batch_first=True)
 
     def forward(self, **input_):

diff --git a/utils.py b/utils.py
@@ -730,8 +730,8 @@ def my_plot(train_acc_list, losses):
         plt.vlines(max_x, min(train_acc_list), max_y, colors='r' if add==0 else 'b', linestyles='dashed')
         plt.hlines(max_y, 0, max_x, colors='r' if add==0 else 'b', linestyles='dashed')
     plt.legend()
-    plt.savefig(os.path.join(os.path.dirname(train_file), 'dev_f1.png'))
+    plt.savefig(os.path.join(os.path.dirname(train_file), f'{MODEL}_dev_f1.png'))
     plt.figure()
     plt.plot(losses)
-    plt.savefig(os.path.join(os.path.dirname(train_file), 'loss.png'))
+    plt.savefig(os.path.join(os.path.dirname(train_file), f'{MODEL}_loss.png'))
 logger = get_logger()