Trademore takes a CSV file of historical stock prices as input, where each row is assumed to be a record of stock price information, and applies time-series prediction models on it to forecast future prices. Under the hood, it leverages sequence-based prediction models from recurrent networks (RNN, LSTM, GRU) to the Transformer (as seen in GPT).
This is not meant to be too heavyweight library with a billion switches and knobs. It is one hackable file, and is mostly intended for educational purposes. PyTorch is the only requirement.
Current implementation follows a few key papers:
- RNN, following Mikolov et al. 2010
- LSTM, following Graves et al. 2014
- GRU, following Kyunghyun Cho et al. 2014
- Transformer, following Vaswani et al. 2017
The included stock_data.csv dataset, as an example, contains historical stock price data for Apple. It looks like:
Date Open High Low Close Adj Close Volume
1980-12-12 0.128348 0.128906 0.128348 0.128348 0.099584 469033600
1980-12-15 0.122210 0.122210 0.121652 0.121652 0.094388 175884800
1980-12-16 0.113281 0.113281 0.112723 0.112723 0.087461 105728000
...
Let's point the script at it:
$ python trademore.py -i stock_data.csv -o stock_model
Training progress and logs and model will all be saved to the working directory stock_model. The default model is a super tiny 200K param transformer; many more training configurations are available - see the argparse and read the code. Training does not require any special hardware, but if you have a GPU then training will be faster. As training progresses the script will print some samples throughout. However, if you'd like to forecast manually, you can use the --forecast-only flag, e.g. in a separate terminal do:
$ python trademore.py -i stock_data.csv -o stock_model --forecast-onlyThis will load the best model so far and print more samples on demand. Remember that predicting stock prices is notoriously difficult, and model predictions should not be used for actual trading without further validation.
Have fun!
MIT