cryptocurrency_prediction's Introduction

Cryptocurrency prediction

Deep tecnical analysis with ANY cryptocurrency

Prerequisites

Miniconda or Anaconda

Setup

Go to configs/vars and edit these lines:

PORTFOLIO_SYMBOLS = [
    'eth',
    'xrp',
    'ltc'
]
TIME_INTERVAL = '1d'
FROM_DATE = '2018-11-01'
TO_DATE = '2019-03-18'

Ubuntu

sudo apt-get install gcc g++ build-essential python-dev python3-dev htop
# make sure you have these installed
conda env create -f UBUNTU_CPU.yml
# create env

Windows

# make sure you have a recent C++ compiler
conda env create -f WINDOWS_CPU.yml
# create env

Mac

conda env create -f MAC_CPU.yml
# create env

python plot_portfolio.py --plot_coin [COIN_NAME]
# e.g.: python plot_portfolio.py --plot_coin eth
# open the /path/to/crytocurrency_prediction/temp-plot.html file

Live demos

Algotrading

registering Env

Add this to ~/miniconda3/envs/crypto_prediction/lib/python3.5/site-packages/gym/envs/__init__.py:

# Custon Env
# ----------------------------------------
register(
    id='Trading-v0',
    entry_point='gym.envs.trading_gym:TradingEnv',
    reward_threshold=2.0,

)

and paste the trading_gym folder inside ~/miniconda3/envs/crypto_prediction/lib/python3.5/site-packages/gym/envs/

Run

rllib train --run PPO --env Trading-v0 --stop '{"timesteps_total": 180000}' --checkpoint-freq 10 --config '{"lr": 1e-5, "num_workers": 2, "observation_filter": "MeanStdFilter"}'

rllib rollout /home/lucas/ray_results/default/PPO_Trading-v0_0_2019-03-26_09-40-05q0q7h143/checkpoint_20/checkpoint-20 --run PPO --env Trading-v0 --steps 1000

Different algorithms compared (mean reward in BTC)

# to keep monitoring while the algo is trainning you can
# run one of these lines in different terminal windows
tensorboard --logdir=~/ray_results
gpustat -i
htop