This project is an exploration into using Keras and Tensorflow to analyse financial data and attempt to make useful predictions. I understand that this may be a fundamentally impossible goal, but nevertheless it's a fun project to explore regression techniques! I've applied it to both traditional stock markets (ASX) and cryptocurrency markets.
Explores:
- Data collection and preparation
- Feature extraction
- Short, medium, and/or long term predictions
- Convolutional layers
- WaveNet (originally by Google)
- LSTM and GRU recurrent neural networks
- Regularization, inc. dropout
- Early stopping, reverting
- Loss functions, metrics, fitness
- Optimizers, learning rates
- Keras Tuner hyperparameter search
- Custom hyperparameter comparisons
This repo is primarily a personal project so it's as clean or user friendly as it would otherwise be.
The configuration file gives an idea of the broad capabilities of the system.
See the example output for a rough idea of what is printed for a single test.
I use venv.
python3 -m venv ./venv
.\venv\Scripts\activate.bat
python3 -m pip install -r requirements.txt
Run Main_CC.py in Jupyter. Using VSCode.
I obtained crypto market historical data from Kraken. See the function ReadKrakenCsv in Crypto_GetData.py.
Dean Reading