Home

Awesome

TRAN

Environment

Python 3.6 & Tensorflow > 1.3

Data

All data, including Stock Historical Sequence Data, Stock Description Document and Stock Relation, are under the data folder. These datasets can be downloaded from: https://drive.google.com/drive/folders/1ybd0XxIpBqJNAax0aioOQHPcEMMv0JoB?usp=sharing.

Stock Historical Sequence Data

Under the 2013-01-01 folder, there are the processed data in NASDAQ and NYSE markets.

Stock Description Document

Under the summary folder, there are the description documents of the stocks and their corresponding companies in NASDAQ and NYSE markets.

Stock Relation

Under the relation folder, there are row relation file storing the sector_industry relations between stocks in NASDAQ and NYSE markets.

Code

Preprocess

sector_industry.py: Generate multi-hot binary vector of industry relation

Training

rank_lstm.py: Train a model of Rank_LSTM
TRAN.py: Train a model of Time-aware Relational Attention Network
TRAN_doc2vec.py: Replace the topic based model with the Doc2vec in the text encoder and then train a model of Time-aware Relational Attention Network

The code is being improved and will be shared in the future.