Quantea - A Machine Learning Trader

This project contains the Quantea framework. Quantea is meant to support Machine Learning based Trading through a standard interface like Sklearn models and allow for ease of backtesting. This project is linked with a Frontend React APP:

Information of how to install this app and run it locally can be found below, with information of how to contribute coming soon!

Table of Contents

• Installing
• Setting Up Quantea
  • Dependencies
• Implementating Your Own Features
  • What Are Features In Trading?
  • Feature Interface
• Implementating Your Own Discretization
  • How to Represent State to an ML Trader
  • Discretization Options
• Writing and Running Tests
• Environment Variables
  • Configuring IEX token

Install

From PyPI with pip (latest stable release):

pip3 install quantea (NOT CURRENTLY AVAILABLE)
pip3 install --index-url https://test.pypi.org/simple --no-dep quantea --upgrade (ONLY TEST AVAILABLE)

From development repository (dev version):

git clone https://github.com/ArcticFaded/Quantea.git
cd quantea
python3 setup.py install

Setting Up Quantea

Quantea relies on MongoDB to cache responses from IEX in order prevent rate limiting API request to IEX cloud while allowing for multiple re-testing sessions.

Dependencies

MongoDB

EXAMPLE USAGE:

from quantea.marketsim.historic_back_trader import HistoricBackTrader
from quantea.technical_indicators.standard_indicators import BollingerBand, EMA, MACD
from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier
from datetime import datetime
from quantea.actions.get_stock_data import get_historical_prices
import numpy as np

start = datetime(2014, 1, 1)
end = datetime.now()

tokens = ['AAPL', 'NVDA']

# example call with fake token (replace with your own)
x = get_historical_prices(start=start, end=end, stocks=tokens, token='your_iex_token_here')

clf = AdaBoostClassifier(n_estimators=2,) #max_depth=2)

trader = HistoricBackTrader(clf, stocks_df=x, train_stock='NVDA', verbose=True)

trader.add_feature(BollingerBand(N_day=26))
trader.add_feature(MACD(N1=26, N2=12))

trader.add_discritizer(lambda x: np.sum(x, axis=1))

tt = trader.train()
testt = trader.test()