Stock prediction based on stock indicators with LSTM and stock-affecting news with BERT WWM

This is the repository of Distributed Computing Research Project supervised by Professor Richard Sinnott.
Language: python Key library: hanlp pytorch pytorch_transformers

Team Members

Tong He
Yao Wang

introduction

There are two models in this repository. Code and result are shown as ipython notebook.

Model I

A hybrid model that combines LSTM and MLP to predict stock.
data_index_2.ipynb
Contains definitions and related methods of the Getdata class for reading and preprocessing data. We provide a copy of processed test data and dev data for the model (train_getdata.pkl,dev_getdata.pkl,test_getdata.pkl).
model_index_2.ipynb
Include the defination of our model and related functions. A sample of trained model is saved in .bin.
evaluate_index.ipynb
Contains predefined function for test and show result.
main_index_v1.ipynb
Main file of Model I，which control the entire process of the project, from data generation to training to evaluation. If '*_getdata.pkl' are given, set "new_data" parameter to False to skip regenerate dataset.

Model II

This is the text classifier for stock related news. bertWWM_classifier.ipynb
Include whole process(data collection,training,evaluation) of this model. Our program is based on PyTorch platform. For the use of the Bert-WWM pre-training model, and the conversion between TensorFlow and PyTorch, please refer to the following two github.
https://github.com/ymcui/Chinese-BERT-wwm
https://github.com/huggingface/transformers
'Construct dataset' cells can be skipped when .csv is given.