Skip to content

Latest commit

 

History

History
35 lines (25 loc) · 1.69 KB

README.md

File metadata and controls

35 lines (25 loc) · 1.69 KB
Raw

Stanford Sentiment Treebank with Multi-Mask

Dataset official web site: https://nlp.stanford.edu/sentiment/

Data Preparation

  • Please download the 6B pre-trained model to ./dataset/glove/. The download address is http://nlp.stanford.edu/data/glove.6B.zip

  • This repo has contained the dataset. Data files checklist for running with default parameters:

      ./dataset/glove/glove.6B.300d.txt
  ./dataset/stanfordSentimentTreebank/datasetSentences.txt
  ./dataset/stanfordSentimentTreebank/datasetSplit.txt
  ./dataset/stanfordSentimentTreebank/dictionary.txt
  ./dataset/stanfordSentimentTreebank/original_rt_snippets.txt
  ./dataset/stanfordSentimentTreebank/sentiment_labels.txt
  ./dataset/stanfordSentimentTreebank/SOStr.txt
  ./dataset/stanfordSentimentTreebank/STree.txt

Training model

After repo cloning and data preparation, Just simply run the code:

cd Proj_SST_mtsa
python3 sst_main.py --network_type mtsa --fine_grained True --model_dir_prefix training --gpu 0

The results will appear at the end of training. We list several frequent use parameters in training. (Please refer to the README.md in repo root for more details).

  • --num_steps: training step;
  • --eval_period: change the evaluation frequency/period.
  • --save_model: [default false] if True, save model to the model ckpt dir;
  • --train_batch_size and --test_batch_size: set to smaller value if GPU memory is limited.
  • --dropout and --wd: dropout keep prob and L2 regularization weight decay.
  • --word_embedding_length and --glove_corpus: word embedding Length and glove model.