def train(date, strategy_id='机器学习1号', working_dir=working_dir, ui_log=None):
date = QA_util_get_real_date(date)
QA_util_log_info('##JOB01 Now Model Init ==== {}'.format(str(date)),
ui_log)
model1 = model()
QA_util_log_info('##JOB02 Now Prepare Date ==== {}'.format(str(date)),
ui_log)
model1.get_data(start=str(int(date[0:4]) - 3) + "-01-01", end=date)
QA_util_log_info('##JOB03 Now Set Target ==== {}'.format(str(date)),
ui_log)
model1.set_target(mark=0.3, type='percent')
QA_util_log_info(
'##JOB04 Now Set Train time range ==== {}'.format(str(date)), ui_log)
model1.set_train_rng(train_start=str(int(date[0:4]) - 3) + "-01-01",
train_end=(datetime.strptime(date, "%Y-%m-%d") -
delta4).strftime('%Y-%m-%d'),
test_start=(datetime.strptime(date, "%Y-%m-%d") -
delta3).strftime('%Y-%m-%d'),
test_end=(datetime.strptime(date, "%Y-%m-%d") -
delta1).strftime('%Y-%m-%d'))
model1.prepare_data()
other_params = {
'learning_rate': 0.1,
'n_estimators': 100,
'max_depth': 5,
'min_child_weight': 1,
'seed': 0,
'subsample': 0.8,
'colsample_bytree': 0.8,
'gamma': 0,
'reg_alpha': 0,
'reg_lambda': 1
}
model1.build_model(other_params)
QA_util_log_info('##JOB05 Now Model Trainnig ==== {}'.format(str(date)),
ui_log)
model1.model_running()
model1.model_check()
QA_util_log_info('##JOB06 Now Save Model ==== {}'.format(str(date)),
ui_log)
important = model1.model_important()
model1.save_model('index', working_dir=working_dir)
QA_util_log_info(
'##JOB06 Now Model Trainning Report ==== {}'.format(str(date)), ui_log)
msg1 = '模型训练日期:{model_date}'.format(model_date=model1.info['date'])
body1 = build_table(pd.DataFrame(model1.info['train_report']), '训练集情况')
body2 = build_table(pd.DataFrame(model1.info['rng_report']), '测试集情况')
body3 = build_table(important.head(50), '特征重要性')
msg = build_email(build_head(), msg1, body1, body2, body3)
send_email('模型训练报告', msg, 'date')
send_actionnotice(strategy_id,
'报告:{}'.format(date),
'模型训练完成,请查收结果',
direction='HOLD',
offset='HOLD',
volume=None)
def job111(trading_date):
if datetime.strptime(trading_date, "%Y-%m-%d").weekday() == 4:
train(trading_date, working_dir=working_dir)
else:
pass
if QA_util_if_trade(trading_date):
predict(trading_date)
else:
predict(QA_util_get_real_date(trading_date))
def subscribe_strategy(
self,
strategy_id: str,
last: int,
today=datetime.date.today(),
cost_coins=10
):
"""订阅一个策略
会扣减你的积分
Arguments:
strategy_id {str} -- [description]
last {int} -- [description]
Keyword Arguments:
today {[type]} -- [description] (default: {datetime.date.today()})
cost_coins {int} -- [description] (default: {10})
"""
if self.coins > cost_coins:
order_id = str(uuid.uuid1())
self._subscribed_strategy[strategy_id] = {
'lasttime':
last,
'start':
str(today),
'strategy_id':
strategy_id,
'end':
QA_util_get_next_day(
QA_util_get_real_date(str(today),
towards=1),
last
),
'status':
'running',
'uuid':
order_id
}
self.coins -= cost_coins
self.coins_history.append(
[
cost_coins,
strategy_id,
str(today),
last,
order_id,
'subscribe'
]
)
return True, order_id
else:
# return QAERROR.
return False, 'Not Enough Coins'
def get(self):
"""[summary]
/fetcher
http://localhost:8010/marketdata/fetcher?code=RB1905&market=future_cn&end=2018-12-01&gap=20&frequence=15min
http://localhost:8010/marketdata/fetcher?code=000001&market=stock_cn&end=2018-12-01&gap=20&frequence=15min
http://localhost:8010/marketdata/fetcher?code=000001,000002&market=stock_cn&end=2018-12-01&gap=20&frequence=realtime&source=tdx
一个统一了多市场的多周期数据接口
param:
code
market
end
gap
frequence
TODO:
source
"""
code = self.get_argument('code', '000001')
market = self.get_argument('market', MARKET_TYPE.STOCK_CN)
end = self.get_argument('end', str(datetime.date.today))
gap = self.get_argument('gap', 50)
frequence = self.get_argument('frequence', FREQUENCE.FIFTEEN_MIN)
start = self.get_argument(
'start', QA_util_get_last_day(QA_util_get_real_date(end),
int(gap)))
source = self.get_argument('source', DATASOURCE.MONGO)
if len(code) > 6:
try:
code = code.split(',')
print(code)
except:
code = code
#print(code, start, end, frequence, market)
res = QA_quotation(code,
start,
end,
frequence,
market,
source=source,
output=OUTPUT_FORMAT.DATASTRUCT)
return self.write({'status': 200, 'result': res.to_json()})
def __init__(self, code, start, end):
super().__init__()
self.frequence = FREQUENCE.DAY
self.market_type = MARKET_TYPE.STOCK_CN
self.backtest_stock_code = code
start = QA_util_get_last_day(
QA_util_get_real_date(start, trade_date_sse, -1), 100)
self.stock_day_data = QAQuery_Advance.QA_fetch_stock_day_adv(
self.backtest_stock_code, start, end)
self.stock_day_data_qfq = self.stock_day_data.to_qfq()
# print(self.stock_day_data)
self.stock_bollinger_bands = QA_indicator_BOLL(
self.stock_day_data_qfq())
# print(len(self.stock_bollinger_bands))
# print(self.stock_bollinger_bands)
self.current_state = 0
def QA_fetch_stock_target(codes, start_date, end_date, type='close'):
if QA_util_if_trade(end_date):
pass
else:
end_date = QA_util_get_real_date(end_date)
end = QA_util_get_next_datetime(end_date, 5)
rng1 = pd.Series(pd.date_range(start_date, end_date,
freq='D')).apply(lambda x: str(x)[0:10])
data = QA.QA_fetch_stock_day_adv(codes, start_date, end)
market = QA.QA_fetch_index_day(['000001'], start_date, end,
format='pd')['close'].reset_index()
market = index_pct(market)[[
'date', 'INDEX_TARGET', 'INDEX_TARGET3', 'INDEX_TARGET4',
'INDEX_TARGET5', 'INDEX_TARGET10'
]]
res1 = data.to_qfq().data
res1.columns = [x + '_qfq' for x in res1.columns]
data = data.data.join(res1).fillna(0).reset_index()
res = data.groupby('code').apply(pct, type=type)[[
'date', 'code', 'PRE_DATE', 'OPEN_MARK', 'PASS_MARK', 'TARGET',
'TARGET3', 'TARGET4', 'TARGET5', 'TARGET10', 'AVG_TARGET'
]]
res = pd.merge(res, market, on='date')
res['date'] = res['date'].apply(lambda x: str(x)[0:10])
res['next_date'] = res['date'].apply(
lambda x: QA_util_get_pre_trade_date(x, -2))
res['PRE_DATE'] = res['PRE_DATE'].apply(lambda x: str(x)[0:10])
res = res.set_index(['date', 'code']).loc[rng1]
res['INDEX_TARGET'] = res['TARGET'] - res['INDEX_TARGET']
res['INDEX_TARGET3'] = res['TARGET3'] - res['INDEX_TARGET3']
res['INDEX_TARGET4'] = res['TARGET4'] - res['INDEX_TARGET4']
res['INDEX_TARGET5'] = res['TARGET5'] - res['INDEX_TARGET5']
res['INDEX_TARGET10'] = res['TARGET10'] - res['INDEX_TARGET10']
for columnname in res.columns:
if res[columnname].dtype == 'float64':
res[columnname] = res[columnname].astype('float16')
if res[columnname].dtype == 'int64':
res[columnname] = res[columnname].astype('int8')
return (res)
def QA_fetch_index_target(codes, start_date, end_date):
if QA_util_if_trade(end_date):
pass
else:
end_date = QA_util_get_real_date(end_date)
end = QA_util_get_next_datetime(end_date, 5)
rng1 = pd.Series(pd.date_range(start_date, end_date,
freq='D')).apply(lambda x: str(x)[0:10])
data = QA.QA_fetch_index_day_adv(codes, start_date,
end).data.fillna(0).reset_index()
res = data.groupby('code').apply(index_pct)[[
'date', 'code', 'INDEX_TARGET', 'INDEX_TARGET3', 'INDEX_TARGET4',
'INDEX_TARGET5', 'INDEX_TARGET10'
]]
res['date'] = res['date'].apply(lambda x: str(x)[0:10])
res = res.set_index(['date', 'code']).loc[rng1]
for columnname in res.columns:
if res[columnname].dtype == 'float64':
res[columnname] = res[columnname].astype('float16')
if res[columnname].dtype == 'int64':
res[columnname] = res[columnname].astype('int8')
return (res)
