a20180502 / lstmrnnstockr

(1) LSTM-RNN stock prices (historical closing precies of S&P500) prediction using keras with tensorflow. (2) Experiments APIs on the network's hyper-parameters are provided through './mmodel/experiment.py'. (3) a website is built using this prediction model as engine with Flask and MySQL.

In this project, I build a LSTM-RNN to predict stock prices using keras with tensorflow. The training data comes from historical closing precies of S&P500. The accuracy measured by Root Mean Square Error (RMSE) is around 0.99. And I did experiments on the network's hyper-parameters such as LSTM cell hidden state size, truncated back propagation length and depth of the network. At last, I build a website using this prediction model as engine with Flask and python.

Dependencies:

python 2.7 pip 9.0.1 flask 0.12.1

tensorflow 0.12.1 keras1.2.1

Run:

Note: Please activate tensorflow virtual env first.

To runthe experiment on LSTM structures:

cd ./model

python experiment.py

To run the web app, first change back to the root

python app.py

Paper & vedio demo

• paper: https://github.com/hualiu01/LSTMRNNStockr/blob/master/paper.pdf
• YouTube vedio demo: https://www.youtube.com/watch?v=haou7Se-RIs&feature=youtu.be