fredmontet / tars Goto Github PK

A crypto trading bot for research and developers

License: MIT License

Makefile 0.12% Jupyter Notebook 97.66% Python 1.93% HTML 0.09% Ruby 0.09% TeX 0.08% Shell 0.01% Smarty 0.01% SCSS 0.01%

tars's Introduction

Tars

Welcome to the repository of Tars; a small trading bot made for research purposes. If you do invest with it, make it with care. ;-)

Prerequisites

The code is structured following the Cookiecutter data science template. Visit https://drivendata.github.io/cookiecutter-data-science/ to become familiar with it.

Dependencies are managed with pipenv, install it following the informations available on https://pipenv.pypa.io/en/latest/.

Getting Started

Once you installed the dependencies with

pipenv install

start your Jupyter Lab with

jupyter lab

and check the getting started notebook in

notebooks/getting-started.ipynb

and try to run it. If it does not work, it's probably because of the dependencies. In this case, you can try to install Tars with pip with

pip install -r requirements.txt

Once it works, take a visit in the notebooks/tutorialsfolder for more.

Have fun!

More information

Go and check the code base in src/tars. It has many comments helping developers like you.

Contribute

Contributions are always welcome, just take an issue and submit pull requests.

Licence

MIT

tars's People

Contributors

Stargazers

Watchers

Forkers

webclinic017

tars's Issues

Strategy backtester

It'd be useful to be able to evaluate the performance of strategies given historical data in an easy way.

This issue relies on the #4 to be executed.

Make a documentation website

It should use Sphinx and the existing notebooks and all comments from the code. The documentation can be minimal in my opinion.

Make utility functions to load historical data

To create models, historical data is a must. The goal of this issue is to create a utility function to load historical data. Historical data can be found here: https://www.cryptodatadownload.com/data/ or with an API here https://polygon.io/ api.

To download historical data can take a long time and therefore, the resulting dataset should be saved somewhere (data folder is probably a good default location).

The location for this function is best in utils/data.py for instance.

Add prerun method

Currently, the strategy run() method is executed at every trade. When it come to using the same model on multiple trades, it's a burden and a method load/train a model before the first run would be required.

Ensemble strategy

The objective of this issue is to provide a feature that allows a user to trade using multiple strategies at the same time on a single portfolio.

For instance, you have 1000 USD in your portfolio and you want to start to use :

BuyAndHold
SequentialInvestment
TrendFollowingMACD

together so that it emulates one person buying, then investing on a periodic basis and trading at the same time.

Make a Telegram bot

When Tars is running, it should be possible to request informations about it's status using a Telegram bot. The commands could be as simple as :

start
stop
get informations

Prediction based strategy

One simple strategy to implement would be to take a trading decision based on a prediction model output. This issue's goal is to implement such a thing with the possibility to import more than one model in the strategy.

For instance, let's define a set of models:

LSTM
Facebook Prophet
Linear Regression

It should be possible to load one, or many of them, in the prediction based strategy so that the trading decision is made given the weighted sum of the prediction of all models.