def test(df, q, periods, actions, pip_mul):
logging.info('Testing: started...')
# initial state
s = getState(df, periods)
# initial action
a = getAction(q, s, 0, actions)
# get reward
r, ticks = getReward(df, a, pip_mul)
# get delta
d = getDelta(q, s, a, r)
return q, r, d, ticks
def test(df, q, periods, actions, pip_mul):
logging.info('Testing: started...')
# initial state
s = getState(df, periods)
# initial action
a = getAction(q, s, 0, actions)
# get reward
r, ticks = getReward(df, a, pip_mul)
# get delta
d = getDelta(q, s, a, r)
return q, r, d, ticks
def train(df, q, alpha, epsilon, periods, actions, pip_mul, std):
logging.info('Training: started...')
d = None
# initial state
s = getState(df, periods)
# initial action
a = getAction(q, s, epsilon, actions)
# get reward
r, ticks = getReward(df, a, pip_mul, std)
# get delta
d = getDelta(q, s, a, r)
# update Q
q = updateQ(q, s, a, d, r, alpha)
return q, r, d, ticks
def train(df, q, alpha, epsilon, periods, actions, pip_mul, std):
logging.info('Training: started...')
d = None
# initial state
s = getState(df, periods)
# initial action
a = getAction(q, s, epsilon, actions)
# get reward
r, ticks = getReward(df, a, pip_mul, std)
# get delta
d = getDelta(q, s, a, r)
# update Q
q = updateQ(q, s, a, d, r, alpha)
return q, r, d, ticks
def predict(df, q, periods, actions, pip_mul, row):
logging.info('Predict: started...')
# state
s = getState(df, periods)
# get top actions
predictions = []
for n in xrange(5):
a, q_sa = getAction(q, s, 0, actions)
a_trade, a_trail = a.split('-')
if a_trade == 'buy':
stop_loss = row['close'] - (float(a_trail) / pip_mul)
else:
stop_loss = row['close'] + (float(a_trail) / pip_mul)
predictions.append('{0} [{1:.4f}] SL:{2:0.4f}'.format(a, q_sa, stop_loss))
logging.info('{0} action = {1}'.format(n, a))
actions.remove(a)
return predictions
def predict(df, q, periods, actions, pip_mul, row):
logging.info('Predict: started...')
# state
s = getState(df, periods)
# get top actions
predictions = []
for n in xrange(5):
a, q_sa = getAction(q, s, 0, actions)
a_trade, a_trail = a.split('-')
if a_trade == 'buy':
stop_loss = row['close'] - (float(a_trail) / pip_mul)
else:
stop_loss = row['close'] + (float(a_trail) / pip_mul)
predictions.append('{0} [{1:.4f}] SL:{2:0.4f}'.format(
a, q_sa, stop_loss))
logging.info('{0} action = {1}'.format(n, a))
actions.remove(a)
return predictions