State Frequency Memory recurrent network for stock price prediction

Author: Liheng Zhang, Date: 08/03/2017

This is the project for the following paper:

Liheng Zhang, Charu Aggarwal, Guo-Jun Qi, Stock Price Prediction via Discovering Multi-Frequency Trading Patterns,
in Proceedings of ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2017), Halifax, Nova Scotia,
Canada, August 13-17, 2017.

Questions about the source codes can be directed to Liheng Zhang at [email protected].

For more applications with SFM, please refer to:

Hao Hu, Guo-Jun Qi. State-Frequency Memory Recurrent Neural Networks, in Proceedings of International Conference
on Machine Learning (ICML 2017), Sydney, Australia, August 6-11, 2017.

Requirements

• Python == 2.7
• Keras == 1.0.1
• Theano == 0.9

Prepare the data

cd dataset; python build_data.py

Test with pretrained model

cd test
python test.py --step=1

The model for n-step prediction is specified with --step. Models for 1-step, 3-step and 5-step prediction are provided.

To visualize the predicted results:

python test --step=1 --visualization=true

Training

cd train
python train.py --step=3 --hidden_dim=50 --freq_dim=10 --niter=4000 --learning_rate=0.01

Note

The codes are expired for Keras >= 2.0.0. Please use the Keras == 1.0.1.