stable/freqtrade/optimize/hyperopt.py

336 lines
11 KiB
Python
Raw Normal View History

# pragma pylint: disable=missing-docstring,W0212,W0603
2017-11-25 00:22:36 +00:00
import json
import logging
import sys
import pickle
import signal
import os
from functools import reduce
from math import exp
from operator import itemgetter
2018-01-10 07:51:36 +00:00
from hyperopt import STATUS_FAIL, STATUS_OK, Trials, fmin, hp, space_eval, tpe
from hyperopt.mongoexp import MongoTrials
from pandas import DataFrame
2018-01-15 21:25:02 +00:00
from freqtrade import main, misc # noqa
from freqtrade import exchange, optimize
2017-11-17 16:54:31 +00:00
from freqtrade.exchange import Bittrex
from freqtrade.misc import load_config
from freqtrade.optimize.backtesting import backtest
2017-12-21 07:31:26 +00:00
from freqtrade.optimize.hyperopt_conf import hyperopt_optimize_conf
from freqtrade.vendor.qtpylib.indicators import crossed_above
2017-11-25 02:28:52 +00:00
# Remove noisy log messages
logging.getLogger('hyperopt.mongoexp').setLevel(logging.WARNING)
2017-12-01 22:21:46 +00:00
logging.getLogger('hyperopt.tpe').setLevel(logging.WARNING)
2017-11-25 02:28:52 +00:00
logger = logging.getLogger(__name__)
# set TARGET_TRADES to suit your number concurrent trades so its realistic to 20days of data
TARGET_TRADES = 600
TOTAL_TRIES = 0
2017-11-25 00:12:44 +00:00
_CURRENT_TRIES = 0
CURRENT_BEST_LOSS = 100
2017-12-01 22:21:46 +00:00
# max average trade duration in minutes
# if eval ends with higher value, we consider it a failed eval
MAX_ACCEPTED_TRADE_DURATION = 300
# this is expexted avg profit * expected trade count
# for example 3.5%, 1100 trades, EXPECTED_MAX_PROFIT = 3.85
# check that the reported Σ% values do not exceed this!
EXPECTED_MAX_PROFIT = 3.0
# Configuration and data used by hyperopt
PROCESSED = None # optimize.preprocess(optimize.load_data())
2017-12-21 07:31:26 +00:00
OPTIMIZE_CONFIG = hyperopt_optimize_conf()
# Hyperopt Trials
2018-01-10 09:50:00 +00:00
TRIALS_FILE = os.path.join('freqtrade', 'optimize', 'hyperopt_trials.pickle')
TRIALS = Trials()
# Monkey patch config
2017-12-16 02:39:47 +00:00
from freqtrade import main # noqa
main._CONF = OPTIMIZE_CONFIG
SPACE = {
2018-01-16 11:31:45 +00:00
'macd_below_zero': hp.choice('macd_below_zero', [
{'enabled': False},
{'enabled': True}
]),
'mfi': hp.choice('mfi', [
{'enabled': False},
{'enabled': True, 'value': hp.quniform('mfi-value', 5, 25, 1)}
]),
'fastd': hp.choice('fastd', [
{'enabled': False},
{'enabled': True, 'value': hp.quniform('fastd-value', 10, 50, 1)}
]),
'adx': hp.choice('adx', [
{'enabled': False},
{'enabled': True, 'value': hp.quniform('adx-value', 15, 50, 1)}
]),
'rsi': hp.choice('rsi', [
{'enabled': False},
{'enabled': True, 'value': hp.quniform('rsi-value', 20, 40, 1)}
]),
'uptrend_long_ema': hp.choice('uptrend_long_ema', [
{'enabled': False},
{'enabled': True}
]),
'uptrend_short_ema': hp.choice('uptrend_short_ema', [
{'enabled': False},
{'enabled': True}
]),
'over_sar': hp.choice('over_sar', [
{'enabled': False},
{'enabled': True}
]),
'green_candle': hp.choice('green_candle', [
{'enabled': False},
{'enabled': True}
]),
'uptrend_sma': hp.choice('uptrend_sma', [
{'enabled': False},
{'enabled': True}
]),
'trigger': hp.choice('trigger', [
{'type': 'lower_bb'},
{'type': 'lower_bb_tema'},
{'type': 'faststoch10'},
{'type': 'ao_cross_zero'},
{'type': 'ema3_cross_ema10'},
{'type': 'macd_cross_signal'},
{'type': 'sar_reversal'},
{'type': 'ht_sine'},
{'type': 'heiken_reversal_bull'},
2018-01-16 16:51:22 +00:00
{'type': 'di_cross'},
]),
'stoploss': hp.uniform('stoploss', -0.5, -0.02),
}
2017-12-16 02:39:47 +00:00
def save_trials(trials, trials_path=TRIALS_FILE):
2018-01-10 09:50:00 +00:00
"""Save hyperopt trials to file"""
logger.info('Saving Trials to \'{}\''.format(trials_path))
pickle.dump(trials, open(trials_path, 'wb'))
def read_trials(trials_path=TRIALS_FILE):
2018-01-10 09:50:00 +00:00
"""Read hyperopt trials file"""
logger.info('Reading Trials from \'{}\''.format(trials_path))
trials = pickle.load(open(trials_path, 'rb'))
os.remove(trials_path)
return trials
2018-01-06 13:53:22 +00:00
def log_trials_result(trials):
vals = json.dumps(trials.best_trial['misc']['vals'], indent=4)
2018-01-06 13:53:22 +00:00
results = trials.best_trial['result']['result']
logger.info('Best result:\n%s\nwith values:\n%s', results, vals)
def log_results(results):
""" log results if it is better than any previous evaluation """
global CURRENT_BEST_LOSS
if results['loss'] < CURRENT_BEST_LOSS:
CURRENT_BEST_LOSS = results['loss']
2018-01-14 11:09:39 +00:00
logger.info('{:5d}/{}: {}. Loss {:.5f}'.format(
results['current_tries'],
results['total_tries'],
2018-01-14 11:09:39 +00:00
results['result'],
results['loss']))
else:
print('.', end='')
sys.stdout.flush()
2017-12-16 02:39:47 +00:00
def calculate_loss(total_profit: float, trade_count: int, trade_duration: float):
2017-12-26 08:08:10 +00:00
""" objective function, returns smaller number for more optimal results """
trade_loss = 1 - 0.25 * exp(-(trade_count - TARGET_TRADES) ** 2 / 10 ** 5.8)
2017-12-26 08:08:10 +00:00
profit_loss = max(0, 1 - total_profit / EXPECTED_MAX_PROFIT)
duration_loss = 0.7 + 0.3 * min(trade_duration / MAX_ACCEPTED_TRADE_DURATION, 1)
return trade_loss + profit_loss + duration_loss
2017-12-26 08:08:10 +00:00
def optimizer(params):
2017-11-25 00:12:44 +00:00
global _CURRENT_TRIES
from freqtrade.optimize import backtesting
backtesting.populate_buy_trend = buy_strategy_generator(params)
results = backtest(OPTIMIZE_CONFIG['stake_amount'], PROCESSED, stoploss=params['stoploss'])
2017-12-26 07:59:38 +00:00
result_explanation = format_results(results)
total_profit = results.profit_percent.sum()
trade_count = len(results.index)
trade_duration = results.duration.mean() * 5
if trade_count == 0 or trade_duration > MAX_ACCEPTED_TRADE_DURATION:
print('.', end='')
return {
'status': STATUS_FAIL,
'loss': float('inf')
}
loss = calculate_loss(total_profit, trade_count, trade_duration)
2017-12-26 08:08:10 +00:00
2017-11-25 00:12:44 +00:00
_CURRENT_TRIES += 1
2017-12-26 07:59:38 +00:00
log_results({
'loss': loss,
'current_tries': _CURRENT_TRIES,
'total_tries': TOTAL_TRIES,
2017-12-26 07:59:38 +00:00
'result': result_explanation,
})
return {
'loss': loss,
'status': STATUS_OK,
2017-12-26 07:59:38 +00:00
'result': result_explanation,
}
def format_results(results: DataFrame):
return ('{:6d} trades. Avg profit {: 5.2f}%. '
2018-01-14 11:10:25 +00:00
'Total profit {: 11.8f} BTC ({:.4f}Σ%). Avg duration {:5.1f} mins.').format(
len(results.index),
results.profit_percent.mean() * 100.0,
results.profit_BTC.sum(),
2018-01-14 11:10:25 +00:00
results.profit_percent.sum(),
results.duration.mean() * 5,
2018-01-14 11:10:25 +00:00
)
def buy_strategy_generator(params):
def populate_buy_trend(dataframe: DataFrame) -> DataFrame:
conditions = []
# GUARDS AND TRENDS
if params['uptrend_long_ema']['enabled']:
conditions.append(dataframe['ema50'] > dataframe['ema100'])
2018-01-16 11:31:45 +00:00
if params['macd_below_zero']['enabled']:
conditions.append(dataframe['macd'] < 0)
2017-11-12 06:45:32 +00:00
if params['uptrend_short_ema']['enabled']:
conditions.append(dataframe['ema5'] > dataframe['ema10'])
if params['mfi']['enabled']:
conditions.append(dataframe['mfi'] < params['mfi']['value'])
if params['fastd']['enabled']:
conditions.append(dataframe['fastd'] < params['fastd']['value'])
if params['adx']['enabled']:
conditions.append(dataframe['adx'] > params['adx']['value'])
2017-10-28 13:21:07 +00:00
if params['rsi']['enabled']:
conditions.append(dataframe['rsi'] < params['rsi']['value'])
if params['over_sar']['enabled']:
conditions.append(dataframe['close'] > dataframe['sar'])
2017-11-11 07:35:04 +00:00
if params['green_candle']['enabled']:
conditions.append(dataframe['close'] > dataframe['open'])
2017-10-20 15:29:38 +00:00
if params['uptrend_sma']['enabled']:
prevsma = dataframe['sma'].shift(1)
conditions.append(dataframe['sma'] > prevsma)
# TRIGGERS
triggers = {
'lower_bb': (dataframe['close'] < dataframe['bb_lowerband']),
'lower_bb_tema': (dataframe['tema'] < dataframe['bb_lowerband']),
'faststoch10': (crossed_above(dataframe['fastd'], 10.0)),
'ao_cross_zero': (crossed_above(dataframe['ao'], 0.0)),
'ema3_cross_ema10': (crossed_above(dataframe['ema3'], dataframe['ema10'])),
'macd_cross_signal': (crossed_above(dataframe['macd'], dataframe['macdsignal'])),
2017-11-11 06:32:35 +00:00
'sar_reversal': (crossed_above(dataframe['close'], dataframe['sar'])),
2017-11-12 07:13:54 +00:00
'ht_sine': (crossed_above(dataframe['htleadsine'], dataframe['htsine'])),
'heiken_reversal_bull': (crossed_above(dataframe['ha_close'], dataframe['ha_open'])) &
(dataframe['ha_low'] == dataframe['ha_open']),
2018-01-16 16:51:22 +00:00
'di_cross': (crossed_above(dataframe['plus_di'], dataframe['minus_di'])),
}
conditions.append(triggers.get(params['trigger']['type']))
dataframe.loc[
reduce(lambda x, y: x & y, conditions),
'buy'] = 1
return dataframe
return populate_buy_trend
2017-10-30 23:36:35 +00:00
def start(args):
global TOTAL_TRIES, PROCESSED, SPACE, TRIALS, _CURRENT_TRIES
2017-11-25 00:12:44 +00:00
TOTAL_TRIES = args.epochs
2017-11-17 16:54:31 +00:00
exchange._API = Bittrex({'key': '', 'secret': ''})
2017-10-30 23:36:35 +00:00
# Initialize logger
logging.basicConfig(
level=args.loglevel,
format='\n%(message)s',
)
logger.info('Using config: %s ...', args.config)
config = load_config(args.config)
pairs = config['exchange']['pair_whitelist']
2018-01-15 21:25:02 +00:00
timerange = misc.parse_timerange(args.timerange)
data = optimize.load_data(args.datadir, pairs=pairs,
ticker_interval=args.ticker_interval,
timerange=timerange)
PROCESSED = optimize.tickerdata_to_dataframe(data)
if args.mongodb:
2017-11-25 02:28:52 +00:00
logger.info('Using mongodb ...')
logger.info('Start scripts/start-mongodb.sh and start-hyperopt-worker.sh manually!')
db_name = 'freqtrade_hyperopt'
TRIALS = MongoTrials('mongo://127.0.0.1:1234/{}/jobs'.format(db_name), exp_key='exp1')
else:
logger.info('Preparing Trials..')
signal.signal(signal.SIGINT, signal_handler)
# read trials file if we have one
if os.path.exists(TRIALS_FILE):
TRIALS = read_trials()
_CURRENT_TRIES = len(TRIALS.results)
TOTAL_TRIES = TOTAL_TRIES + _CURRENT_TRIES
logger.info(
'Continuing with trials. Current: {}, Total: {}'
.format(_CURRENT_TRIES, TOTAL_TRIES))
2018-01-08 01:45:31 +00:00
try:
best_parameters = fmin(
fn=optimizer,
space=SPACE,
algo=tpe.suggest,
max_evals=TOTAL_TRIES,
trials=TRIALS
2018-01-08 01:45:31 +00:00
)
results = sorted(TRIALS.results, key=itemgetter('loss'))
2018-01-08 01:45:31 +00:00
best_result = results[0]['result']
except ValueError:
best_parameters = {}
best_result = 'Sorry, Hyperopt was not able to find good parameters. Please ' \
'try with more epochs (param: -e).'
# Improve best parameter logging display
2018-01-08 01:45:31 +00:00
if best_parameters:
best_parameters = space_eval(SPACE, best_parameters)
2018-01-08 01:45:31 +00:00
logger.info('Best parameters:\n%s', json.dumps(best_parameters, indent=4))
logger.info('Best Result:\n%s', best_result)
# Store trials result to file to resume next time
save_trials(TRIALS)
def signal_handler(sig, frame):
2018-01-10 09:50:00 +00:00
"""Hyperopt SIGINT handler"""
logger.info('Hyperopt received {}'.format(signal.Signals(sig).name))
save_trials(TRIALS)
2018-01-06 13:53:22 +00:00
log_trials_result(TRIALS)
sys.exit(0)