Sentiment analysis in PyTorch, algorithm implemented as follows:

• LSTM
• LSTM with glove-pretrain embedding weights
• Bidirectional LSTM
• DistilBert

• Download directly: https://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz
• HuggingFace dataset: https://huggingface.co/datasets/imdb

from datasets import load_dataset
load_dataset('imdb').with_format('torch') # for pytorch format

Apply pre-train glove pre-train word vector for LSTM embedding weight

• DataSource
  • github: https://github.com/stanfordnlp/GloVe
  • Kaggle: https://www.kaggle.com/datasets/anindya2906/glove6b
• Preprocess: transform gloveweight to gensim format

unzip glove.6B.zip
python -m gensim.scripts.glove2word2vec --input glove.6B.100d.txt --output glove.6B.100d.w2vformat.txt

## load glove pre-train weight
import gensim
word_vec = gensim.models.KeyedVectors.load_word2vec_format('glove.6B.100d.w2vformat.txt', binary=False, encoding='utf-8')

# --m: model type(default/lstm: LSTM, glove-lstm: LSTM with glove-pretrain embedding weights, 
#                 bi-lstm:Bidirectional LSTM, distilbert:DistilBert)
# --e: epoch (default:2)
python main.py --m "glove-lstm" --e 3

Use gradio to demo model result, please click link to give it a try!