miaohua1982 / sentimentimdb Goto Github PK

A demo for NLP sentiment by dataset imdb using pytorch framework
Rightnow, I just realize 7 models: lstm, gru, fasttext, cnn, bert, adbert, transformer_enc


#training command line for fasttext with max seq length=512
python main.py --train --model fast --model_path model_storage/model_fast.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer spacy --embedding_size 100 -bs 64 -d /root/myfavor/nlp/.data/imdb/aclImdb
#testing command line for fasttext(0.877)
python main.py --test --model fast --load_model --model_path model_storage/model_fast.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer spacy --embedding_size 100 -bs 64 -d /root/myfavor/nlp/.data/imdb/aclImdb

#with bert weights it can be to 0.884
python main.py --train --model fast --model_path model_storage/model_fast.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer bert --embedding_size 768 -bs 64 -d /root/myfavor/nlp/.data/imdb/aclImdb
python main.py --test --model fast --load_model --model_path model_storage/model_fast.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer bert --embedding_size 768 -bs 64 -d /root/myfavor/nlp/.data/imdb/aclImdb

#training command line for cnn classifier with max seq length=512
python main.py --train --model cnn --model_path model_storage/model_cnn.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer spacy --embedding_size 100 --ff_hidden 256 -bs 64 -e 3 -d /root/data/aclImdb
#testing command line for cnn classifier(0.894)
python main.py --test --model cnn --load_model --model_path model_storage/model_cnn.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer spacy --embedding_size 100 --ff_hidden 256 -bs 64 -d /root/data/aclImdb

#training command line lstm
python main.py --train --model lstm --model_path model_storage/model_lstm.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer spacy --ff_hidden 256 --drop_out 0.5 -bs 128 --rnn_layers 2 -d /root/data/aclImdb --embedding_size 100
#testing command line for lstm(0.858)
python main.py --test --model lstm --load_model --model_path model_storage/model_lstm.pth --tokenizer spacy --embedding_size 100 --ff_hidden 256 -bs 64 --rnn_layers 2 -d /root/data/aclImdb

#training command line for gru
python main.py --train --model gru --model_path model_storage/model_gru.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer spacy --ff_hidden 256 --drop_out 0.5 -bs 128 --rnn_layers 2 -d /root/data/aclImdb --embedding_size 100
#testing command line for gru(0.855)
python main.py --test --model gru --load_model --model_path model_storage/model_gru.pth --tokenizer spacy --embedding_size 100 --ff_hidden 256 -bs 64 --rnn_layers 2 -d /root/data/aclImdb


#use bert tokenizer & pretrained weights(with max seq lengths=512, batch_size=128)
python main.py --train --model bert --model_path model_storage/model_bert.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer bert --embedding_size 768 --ff_hidden 256 --drop_out 0.25 -bs 128 --rnn_layers 2 -d /root/myfavor/nlp/.data -e 2
#testing command for bert(0.882)
python main.py --test --model bert --model_path model_storage/model_bert.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer bert --embedding_size 768 --ff_hidden 256 --drop_out 0.25 -bs 128 --rnn_layers 2 -d /root/myfavor/nlp/.data

#use torchtext for adbert
python main.py --train --torchtext --model adbert --model_path model_storage/model_adbert.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer adbert --embedding_size 768 --ff_hidden 256 --drop_out 0.25 -bs 128 --rnn_layers 2 -d /root/myfavor/nlp/.data -e 3
#testing command for adbert(0.922)
python main.py --test --torchtext --model adbert --load_model --model_path model_storage/model_adbert.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer adbert --embedding_size 768 --ff_hidden 256 --drop_out 0.25 -bs 128 --rnn_layers 2 -d /root/myfavor/nlp/.data

#training command line for transformer_enc
python main.py --train --model transformer_enc --model_path model_storage/model_transformer_enc.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer spacy --embedding_size 100 --drop_out 0.1 --cuda -e 4 --enc_layers 6 -d /root/myfavor/nlp/.data
python main.py --test --model transformer_enc --model_path model_storage/model_transformer_enc.pth --tokenizer_dump_path model_storage/tokenizer_file.dmp --tokenizer spacy --embedding_size 100 --drop_out 0.1 --cuda -e 4 --enc_layers 6 -d /root/myfavor/nlp/.data