Merge pull request #4 from lyhuang18/lyhuang-reproduction

update README.md

reproduction/text_classification/README.md

@@ -1,22 +1,28 @@
 # text_classification任务模型复现
 这里使用fastNLP复现以下模型：
+
 char_cnn :论文链接[Character-level Convolutional Networks for Text Classification](https://arxiv.org/pdf/1509.01626v3.pdf)
+
 dpcnn:论文链接[Deep Pyramid Convolutional Neural Networks for TextCategorization](https://ai.tencent.com/ailab/media/publications/ACL3-Brady.pdf)
+
 HAN:论文链接[Hierarchical Attention Networks for Document Classification](https://www.cs.cmu.edu/~diyiy/docs/naacl16.pdf)
+
+LSTM+self_attention:论文链接[A Structured Self-attentive Sentence Embedding](<https://arxiv.org/pdf/1703.03130.pdf>)
+
+AWD-LSTM:论文链接[Regularizing and Optimizing LSTM Language Models](<https://arxiv.org/pdf/1708.02182.pdf>)
+
 #待补充
-awd_lstm:
-lstm_self_attention(BCN?):
-awd-sltm:
 
 # 数据集及复现结果汇总
 
 使用fastNLP复现的结果vs论文汇报结果(/前为fastNLP实现，后面为论文报道,-表示论文没有在该数据集上列出结果)
 
-model name | yelp_p | sst-2|IMDB| 
-:---: | :---: | :---: | :---: 
-char_cnn | 93.80/95.12 | - |- |
-dpcnn | 95.50/97.36 | - |- |
-HAN |- | - |-|
-BCN| - |- |-|
-awd-lstm| - |- |-|
+model name | yelp_p | yelp_f | sst-2|IMDB
+:---: | :---: | :---: | :---: |----- 
+char_cnn | 93.80/95.12 | - | - |- 
+dpcnn | 95.50/97.36 | - | - |- 
+HAN |- | - | - |-
+LSTM| 95.74/- |- |- |88.52/-
+AWD-LSTM| 95.96/- |- |- |88.91/-
+LSTM+self_attention| 96.34/- | - | - |89.53/-
 

reproduction/text_classification/train_awdlstm.py

@@ -8,9 +8,7 @@ os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches'
 
 import torch.nn as nn
 
-from data.SSTLoader import SSTLoader
 from data.IMDBLoader import IMDBLoader
-from data.yelpLoader import yelpLoader
 from fastNLP.modules.encoder.embedding import StaticEmbedding
 from model.awd_lstm import AWDLSTMSentiment
 
@@ -35,24 +33,15 @@ class Config():
 
     task_name = "IMDB"
     datapath={"train":"IMDB_data/train.csv", "test":"IMDB_data/test.csv"}
-    load_model_path="./result_IMDB/best_BiLSTM_SELF_ATTENTION_acc_2019-07-07-04-16-51"
     save_model_path="./result_IMDB_test/"
-opt=Config
 
+opt=Config()
 
-# load data
-dataloaders = {
-    "IMDB":IMDBLoader(),
-    "YELP":yelpLoader(),
-    "SST-5":SSTLoader(subtree=True,fine_grained=True),
-    "SST-3":SSTLoader(subtree=True,fine_grained=False)
-}
-
-if opt.task_name not in ["IMDB", "YELP", "SST-5", "SST-3"]:
-    raise ValueError("task name must in ['IMDB', 'YELP, 'SST-5', 'SST-3']")
 
-dataloader = dataloaders[opt.task_name]
+# load data
+dataloader=IMDBLoader()
 datainfo=dataloader.process(opt.datapath)
+
 # print(datainfo.datasets["train"])
 # print(datainfo)
 
@@ -71,32 +60,10 @@ optimizer= Adam([param for param in model.parameters() if param.requires_grad==T
 
 def train(datainfo, model, optimizer, loss, metrics, opt):
     trainer = Trainer(datainfo.datasets['train'], model, optimizer=optimizer, loss=loss,
-                        metrics=metrics, dev_data=datainfo.datasets['dev'], device=0, check_code_level=-1,
+                        metrics=metrics, dev_data=datainfo.datasets['test'], device=0, check_code_level=-1,
                         n_epochs=opt.train_epoch, save_path=opt.save_model_path)
     trainer.train()
 
 
-def test(datainfo, metrics, opt):
-    # load model
-    model = ModelLoader.load_pytorch_model(opt.load_model_path)
-    print("model loaded!")
-
-    # Tester
-    tester = Tester(datainfo.datasets['test'], model, metrics, batch_size=4, device=0)
-    acc = tester.test()
-    print("acc=",acc)
-
-
-
-parser = argparse.ArgumentParser()
-parser.add_argument('--mode', required=True, dest="mode",help='set the model\'s model')
-
-
-args = parser.parse_args()
-if args.mode == 'train':
+if __name__ == "__main__":
     train(datainfo, model, optimizer, loss, metrics, opt)
-elif args.mode == 'test':
-    test(datainfo, metrics, opt)
-else:
-    print('no mode specified for model!')
-    parser.print_help()

reproduction/text_classification/train_lstm.py

@@ -6,9 +6,7 @@ os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches'
 
 import torch.nn as nn
 
-from data.SSTLoader import SSTLoader
 from data.IMDBLoader import IMDBLoader
-from data.yelpLoader import yelpLoader
 from fastNLP.modules.encoder.embedding import StaticEmbedding
 from model.lstm import BiLSTMSentiment
 
@@ -32,24 +30,15 @@ class Config():
 
     task_name = "IMDB"
     datapath={"train":"IMDB_data/train.csv", "test":"IMDB_data/test.csv"}
-    load_model_path="./result_IMDB/best_BiLSTM_SELF_ATTENTION_acc_2019-07-07-04-16-51"
     save_model_path="./result_IMDB_test/"
-opt=Config
 
+opt=Config()
 
-# load data
-dataloaders = {
-    "IMDB":IMDBLoader(),
-    "YELP":yelpLoader(),
-    "SST-5":SSTLoader(subtree=True,fine_grained=True),
-    "SST-3":SSTLoader(subtree=True,fine_grained=False)
-}
-
-if opt.task_name not in ["IMDB", "YELP", "SST-5", "SST-3"]:
-    raise ValueError("task name must in ['IMDB', 'YELP, 'SST-5', 'SST-3']")
 
-dataloader = dataloaders[opt.task_name]
+# load data
+dataloader=IMDBLoader()
 datainfo=dataloader.process(opt.datapath)
+
 # print(datainfo.datasets["train"])
 # print(datainfo)
 
@@ -68,32 +57,10 @@ optimizer= Adam([param for param in model.parameters() if param.requires_grad==T
 
 def train(datainfo, model, optimizer, loss, metrics, opt):
     trainer = Trainer(datainfo.datasets['train'], model, optimizer=optimizer, loss=loss,
-                        metrics=metrics, dev_data=datainfo.datasets['dev'], device=0, check_code_level=-1,
+                        metrics=metrics, dev_data=datainfo.datasets['test'], device=0, check_code_level=-1,
                         n_epochs=opt.train_epoch, save_path=opt.save_model_path)
     trainer.train()
 
 
-def test(datainfo, metrics, opt):
-    # load model
-    model = ModelLoader.load_pytorch_model(opt.load_model_path)
-    print("model loaded!")
-
-    # Tester
-    tester = Tester(datainfo.datasets['test'], model, metrics, batch_size=4, device=0)
-    acc = tester.test()
-    print("acc=",acc)
-
-
-
-parser = argparse.ArgumentParser()
-parser.add_argument('--mode', required=True, dest="mode",help='set the model\'s model')
-
-
-args = parser.parse_args()
-if args.mode == 'train':
-    train(datainfo, model, optimizer, loss, metrics, opt)
-elif args.mode == 'test':
-    test(datainfo, metrics, opt)
-else:
-    print('no mode specified for model!')
-    parser.print_help()
+if __name__ == "__main__":
+    train(datainfo, model, optimizer, loss, metrics, opt)

reproduction/text_classification/train_lstm_att.py

@@ -6,9 +6,7 @@ os.environ['FASTNLP_CACHE_DIR'] = '/remote-home/hyan01/fastnlp_caches'
 
 import torch.nn as nn
 
-from data.SSTLoader import SSTLoader
 from data.IMDBLoader import IMDBLoader
-from data.yelpLoader import yelpLoader
 from fastNLP.modules.encoder.embedding import StaticEmbedding
 from model.lstm_self_attention import BiLSTM_SELF_ATTENTION
 
@@ -34,24 +32,15 @@ class Config():
 
     task_name = "IMDB"
     datapath={"train":"IMDB_data/train.csv", "test":"IMDB_data/test.csv"}
-    load_model_path="./result_IMDB/best_BiLSTM_SELF_ATTENTION_acc_2019-07-07-04-16-51"
     save_model_path="./result_IMDB_test/"
-opt=Config
 
+opt=Config()
 
-# load data
-dataloaders = {
-    "IMDB":IMDBLoader(),
-    "YELP":yelpLoader(),
-    "SST-5":SSTLoader(subtree=True,fine_grained=True),
-    "SST-3":SSTLoader(subtree=True,fine_grained=False)
-}
-
-if opt.task_name not in ["IMDB", "YELP", "SST-5", "SST-3"]:
-    raise ValueError("task name must in ['IMDB', 'YELP, 'SST-5', 'SST-3']")
 
-dataloader = dataloaders[opt.task_name]
+# load data
+dataloader=IMDBLoader()
 datainfo=dataloader.process(opt.datapath)
+
 # print(datainfo.datasets["train"])
 # print(datainfo)
 
@@ -70,32 +59,10 @@ optimizer= Adam([param for param in model.parameters() if param.requires_grad==T
 
 def train(datainfo, model, optimizer, loss, metrics, opt):
     trainer = Trainer(datainfo.datasets['train'], model, optimizer=optimizer, loss=loss,
-                        metrics=metrics, dev_data=datainfo.datasets['dev'], device=0, check_code_level=-1,
+                        metrics=metrics, dev_data=datainfo.datasets['test'], device=0, check_code_level=-1,
                         n_epochs=opt.train_epoch, save_path=opt.save_model_path)
     trainer.train()
 
 
-def test(datainfo, metrics, opt):
-    # load model
-    model = ModelLoader.load_pytorch_model(opt.load_model_path)
-    print("model loaded!")
-
-    # Tester
-    tester = Tester(datainfo.datasets['test'], model, metrics, batch_size=4, device=0)
-    acc = tester.test()
-    print("acc=",acc)
-
-
-
-parser = argparse.ArgumentParser()
-parser.add_argument('--mode', required=True, dest="mode",help='set the model\'s model')
-
-
-args = parser.parse_args()
-if args.mode == 'train':
+if __name__ == "__main__":
     train(datainfo, model, optimizer, loss, metrics, opt)
-elif args.mode == 'test':
-    test(datainfo, metrics, opt)
-else:
-    print('no mode specified for model!')
-    parser.print_help()