This repository explores exploiting Bitcoin price patterns using deep learning, targeting researchers and traders interested in AI-driven financial analysis. It aims to predict price movements by training neural networks on historical price data visualized as images.

How It Works

The project converts Bitcoin tick data into OHLC (Open, High, Low, Close) images, treating price patterns as visual data. It then trains convolutional neural networks (CNNs), starting with AlexNet, on these images to predict price direction (UP/DOWN) or magnitude. This approach mimics human visual pattern recognition for trading decisions.

Quick Start & Requirements

• Install: git clone https://github.com/philipperemy/deep-learning-bitcoin.git
• Run: cd deep-learning-bitcoin, then ./data_download.sh and python3 data_generator.py.
• Prerequisites: Python 3, Caffe framework, NVIDIA DIGITS, and potentially CUDA-enabled GPU for training.
• Docker: docker build -t dlb . and docker run -it --name dlb -v $PWD:/app dlb /bin/bash.
• Documentation: Stack Overflow TclError solution for display issues.

Highlighted Details

• Achieved 70% accuracy predicting UP/DOWN price movements on 20,000 samples using AlexNet.
• Data processing includes converting tick data to 5-minute OHLC representations.
• Future plans include training on larger datasets, using more complex CNNs (LeNet), and integrating volume data and attention mechanisms.

Maintenance & Community

The repository appears to be a personal project with no explicit mention of active maintenance, community channels, or notable contributors.

Licensing & Compatibility

The repository does not explicitly state a license.

Limitations & Caveats

The project is in an early stage, with initial results based on a relatively small dataset (20,000 samples). Key planned features like training on larger datasets and integrating advanced CNNs are not yet implemented.