Cryptocurrency algorithmic trading using neural networks

It is my first project of more extensive scope. This is where I acknowledged my interest in both Machine learning and algorythmic trading. This project consists of a rather simple LSTM recurrent neural network builder (using Keras). The goal of the neural network is to predict cryptocurenncy price movements - preferrably short term. And it succeeded to do so.

How do I know how well the model will perform when utilized for trading in the future? I don't. I can only look on historical data and try to determine whether it has potential to be effective and make money regardless of market behavior. A big part of my work on this project was to construct tools that best describe model's performance and robustness. Using plotly I made charts that visualise model's prediction pattern and it's accuracy. Also, I made a backtesting simulation of a simple strategies that can be based on model's prediction - and visualisation of how profitable they would be over time. Lastly, I implemented a simple trading bot that can preprocess data, run predictions and send orders to the exchange (Binance) via API.

The project's code and overall structure may seem messy. It is not a suprise since it's my first work of that kind and I was just a high school student. I've learnt a lot since and I will put my work into imporving the structure, code readability and speed along with ease of use . To be honest, while I'm still passionate about machine learning and its applications for algorithmic trading, I probably won't be maintaining and updating this particular project.

Project structure:

• Data processing and model training:

  • getdata.py - tool used to download historical data from the exchange
  • new_preprocess.py - front to data preprocessing
  • compile.py - model training
  • vars.py - some constant variables
  • utils.py - utilities
  • data_processing.py - data processing functions - used by new_preprocess.py, dist_acc_graph.py, predictions_graph.py

• Visualisation:

  • dist_acc_graph.py - model's distribution of accuracy chart
  • predictions_graph.py - model's chart of predictions plotted along price chart

• Backtesting:

  • backtesting/brain.py - place to set a strategy and run a backtest
  • backtesting/heart.py - 'inner' processies of backtest
  • backtesting/statistics.py - analisys and saving trades
  • backtesting/chart.py - visualisation of strategy's performance over time
  • backtesting/chart_all.py - visualisation of multiple strategies' performance over time

• Realtime - basically a trading bot for binance (probably messy and built for deprecated api :/)

  • realtime/BRAIN.py - place to set a strategy and run a bot
  • realtime/HEART.py - account endpoint, placing and monitoring orders
  • realtime/GENERAL.py - ulitities

Charts

dist_acc_graph.py

Bit more thorought analysis of model's accuracy based on how certain the model is of its prediction

predictions_graph.py

Visualisation of model's predictions along with the price chart at the time.

backtesting/chart.py

Profitability chart of a strategy with paremeters:

    LEVERAGE = 7
    ORDER_SIZE = 0.1
    PYRAMID_MAX = 1
    THRESHOLD = 0.08

backtesting/chart_all.py

Comparison on strategies with paremeters:

    thresholds = [0.04, 0.06, 0.08, 0.1, 0.12, 0.16]
    leverages = [7]
    order_sizes = [0.1]
    pyramid_maxes = [1]