Text Classification Benchmark

A Benchmark of Text Classification in PyTorch

Motivation

We are trying to build a Benchmark for Text Classification including

Many Text Classification DataSet, including Sentiment/Topic Classfication, popular language(e.g. English and Chinese). Meanwhile, a basic word embedding is provided.

Implment many popular and state-of-art Models, especially in deep neural network.

Have done

We have done some dataset and models

Dataset done

• IMDB
• SST
• Trec

Models done

• FastText
• BasicCNN (KimCNN,MultiLayerCNN, Multi-perspective CNN)
• InceptionCNN
• LSTM (BILSTM, StackLSTM)
• LSTM with Attention (Self Attention / Quantum Attention)
• Hybrids between CNN and RNN (RCNN, C-LSTM)
• Transformer - Attention is all you need
• ConS2S
• Capsule
• Quantum-inspired NN

Libary

You should have install these librarys

python3
torch
torchtext (optional)

Dataset

Dataset will be automatically configured in current path, or download manually your data in Dataset, step-by step.

including

Glove embeding
Sentiment classfication dataset IMDB

usage

Run in default setting

python main.py

CNN

python main.py --model cnn

LSTM

python main.py --model lstm

Road Map

• Data preprossing framework
• Models modules
• Loss, Estimator and hyper-paramter tuning.
• Test modules
• More Dataset
• More models

Organisation of the repository

The core of this repository is models and dataset.

• dataloader/: loading all dataset such as IMDB, SST

• models/: creating all models such as FastText, LSTM,CNN,Capsule,QuantumCNN ,Multi-Head Attention

• opts.py: Parameter and config info.

• utils.py: tools.

• dataHelper: data helper

Contributor

• @Allenzhai
• @JaredWei
• @AlexMeng
• @Lilianwang
• @ZhanSu
• @Wabywang

Welcome your issues and contribution!!!