Trading-Gym

Trading-Gym is a trading environment base on Gym. For those who want to custom everything.

install

$ pip install trading-gym

Creating features with ta-lib is suggested, that will improve the performance of agent and make it easy to learn. You should install ta-lib before it. Take Ubuntu x64 for example.

$ wget http://prdownloads.sourceforge.net/ta-lib/ta-lib-0.4.0-src.tar.gz 
$ tar -zxvf ta-lib-0.4.0-src.tar.gz
$ cd ta-lib/
$ ./configure --prefix=$PREFIX
$ make install
$ export TALIBRARYPATH=$PREFIX/lib $ export TAINCLUDEPATH=$PREFIX/include
$ pip install TA-Lib

See more.

Examples

quick start

from trading_gym.env import TradeEnv
import random
env = TradeEnv(datapath='./data/testexchange.json') done = False obs = env.reset() for i in range(500):     action = random.sample([0, 1, 2], 1)[0]     obs, reward, done, info = env.step(action)     env.render()     if done:         break

A sample train with stable-baselines

from trading_gym.env import TradeEnv
from stablebaselines.common.vecenv import DummyVecEnv
from stable_baselines import DQN
from stable_baselines.deepq.policies import MlpPolicy
datapath = './data/fakesin_data.json' env = TradeEnv(datapath=datapath, unit=50000, datakwargs={'useta': True}) env = DummyVecEnv([lambda: env])
model = DQN(MlpPolicy, env, verbose=2, learning_rate=1e-5) model.learn(200000)
obs = env.reset() for i in range(8000):     action, _states = model.predict(obs)     obs, rewards, done, info = env.step(action)     env.render()     if done:         break

input format

[
    {
        "open": 10.0,
        "close": 10.0,
        "high": 10.0,
        "low": 10.0,
        "volume": 10.0,
        "date": "2019-01-01 09:59"
    },
    {
        "open": 10.1,
        "close": 10.1,
        "high": 10.1,
        "low": 10.1,
        "volume": 10.1,
        "date": "2019-01-01 10:00"
    }
]

actions

| Action | Value | | ------ | ----- | | PUT | 0 | | HOLD | 1 | | PUSH | 2 |

observation

• native obs: shape=(*, 51, 6), return 51 history data with OCHL

env = TradeEnv(datapath=datapath)

• obs with ta: shape=(*, 10), return obs using talib.
• - default feature: ['ema', 'wma', 'sma', 'sar', 'apo', 'macd', 'macdsignal', 'macdhist', 'adosc', 'obv']

env = TradeEnv(datapath=datapath, datakwargs={'useta': True})

Custom

custom obs

def customobsfeatures_func(history, info):
    close = [obs.close for obs in history]
    return close
env = TradeEnv(datapath=datapath,                getobsfeaturesfunc=customobsfeaturesfunc,                ops_shape=(1))

custom reward

def customrewardfunc(exchange):
    return exchange.profit
env = TradeEnv(datapath=datapath,                getrewardfunc=customrewardfunc)

Param exchange is entity of Exchange

Reward

• reward = fixed_profit
• profit = fixedprofit + floatingprofit
• floatingprofit = (latestprice - avg_price) * unit
• unit = int(nav / buyinprice)
• avgprice = ((buyin_price * unit) + charge) / unit
• fixedprofit = SUM([every floatingprofit after close position])