Check my blog post "Predict Stock Prices Using RNN": Part 1 and Part 2 for the tutorial associated.
One thing I would like to emphasize that because my motivation is more on demonstrating how to build and train an RNN model in Tensorflow and less on solve the stock prediction problem, I didn't try too hard on improving the prediction outcomes. You are more than welcome to take this repo as a reference point and add more stock prediction related ideas to improve it. Enjoy.
- Make sure
tensorflow
has been installed. - First download the full S&P 500 data from Yahoo! Finance ^GSPC (click the "Historical Data" tab and select the max time period). And save the .csv file to
data/SP500.csv
. - Run
python data_fetcher.py
to download the prices of individual stocks in S & P 500, each saved todata/{{stock_abbreviation}}.csv
. (NOTE: Google Finance API returns the prices for 4000 days maximum. If you are curious about the data, try modifyingdata_fetcher.py
code to send multiple queries for a stock. Here is the data archive (stock-data-lilianweng.tar.gz) of stock prices I stored up to Jul, 2017. Please unzip this file to replace the "data" folder in the repo for test runs.) - Run
python main.py --help
to check the available command line args. - Run
python main.py
to train the model.
Examples
- Train a model only on SP500.csv; no embedding
python main.py --stock_symbol=SP500 --train --input_size=1 --lstm_size=128 --max_epoch=50
- Train a model on 100 stocks; with embedding of size 8
python main.py --stock_count=100 --train --input_size=1 --lstm_size=128 --max_epoch=50 --embed_size=8
- Start your Tensorboard
cd stock-rnn
mkdir logs
tensorboard --logdir ./logs --port 1234 --debug
My python environment: Python version == 2.7
BeautifulSoup==3.2.1
numpy==1.13.1
pandas==0.16.2
scikit-learn==0.16.1
scipy==0.19.1
tensorflow==1.2.1
urllib3==1.8