def get_data_for_selection(self, start_date, end_date):
curror = BaseModel('novel_Feature').query(
sql=dict(stock_code={'$in': model.li},
date={
'$gte': start_date,
'$lte': end_date
}))
if curror.count():
data = pd.DataFrame(list(curror))
data = data.loc[:, model.names]
data = data.replace(to_replace=np.Infinity, value=np.NaN).dropna()
# data['change_r_next16'] = data.change_r_next16.map(lambda x: 1 if x > 0 else 0)
for cl in model.names:
v = data[cl].iloc[0]
# tp=type(v)
# if type(v).__name__ !='bool':
# data[cl] = data.sort_values(by=[cl], ascending=False).index
# # temp = data[cl].mean()
# data[cl] = data[cl].map(lambda x:1 if x>len(data)//2 else 0)
# pass
y = data.loc[:, [model.label]]
y['sort'] = y.sort_values(by=[model.label], ascending=False).index
y = y.drop([model.label], axis=1)
y['sort'] = y.sort.map(lambda x: 1 if x > len(y) / 2 else 0)
X = data.drop([model.label, 'date', '_id'], axis=1)
return X, y
else:
return [], []
def get_data(self, start_date, end_date):
print(start_date)
curror = BaseModel('features_index_day').query(
sql=dict(stock_code={'$in': model.li},
date={
'$gte': start_date,
'$lte': end_date
}))
if curror.count():
data = pd.DataFrame(list(curror))
data = data.loc[:, model.names + [model.label]]
data = data.replace(to_replace=np.Infinity, value=np.NaN).dropna()
# data['change_r_next16'] = data.change_r_next16.map(lambda x: 1 if x > 0 else 0)
for cl in model.names:
v = data[cl].iloc[0]
# if type(v).__name__ !='bool':
# data[cl] = data.sort_values(by=[cl], ascending=False).index
# temp = data[cl].mean()
# data[cl] = data[cl].map(lambda x:1 if x>len(data)//2 else 0)
# pass
y = data.loc[:, [model.label]]
y['sort'] = y.sort_values(by=[model.label], ascending=False).index
y = y.drop([model.label], axis=1)
y['sort'] = y.sort.map(lambda x: self.classify(x))
# y['sort'] = y.sort.map(lambda x:1 if x>len(y)/2 else 0)
X = data.drop([model.label], axis=1)
# X = data.loc[:, model.names]
# X_test = data_test.loc[:, model.names]
# y_test = data_test.loc[:, ['change_r_next16']]
return X, y
else:
return pd.DataFrame([]), pd.DataFrame([])
def min60(time):
min60_times = [1030, 1130, 1400, 1500]
if time in min60_times:
temp = time // 100 * 60 + time % 100 - 60
temp = temp // 60 * 100 + temp % 60
print(temp)
sql = {'time': {'$gt': temp, '$lte': time}, 'date': Calendar.today()}
curror = BaseModel('real_kline_min5').query(sql=sql)
if curror.count():
data = pd.DataFrame(list(curror))
data = data.sort_values(by=['time'], ascending=True)
data = data.groupby(by=['stock_code'], as_index=False).agg({
'volume':
'sum',
'amount':
'sum',
'open':
'first',
'close':
'last',
'high':
'max',
'low':
'min'
})
data['time'] = time
data['date'] = Calendar.today()
BaseModel('real_kline_min60').insert_batch(
data.to_dict(orient='records'))
print('min60 ok')
# min60(time=1030)
if __name__ == '__main__':
# clf = DecisionTreeClassifier()
# clf = LogisticRegression()
table = 'features_index_day'
clf = RandomForestClassifier(n_estimators=200, n_jobs=2)
label = 'change_r_next2'
curror = BaseModel(table).query(
sql={
'date': {
'$gte': dt.datetime(2018, 1, 5),
'$lte': dt.datetime(2018, 7, 25)
}
})
print(curror.count())
if curror.count():
data = pd.DataFrame(list(curror))
data = data.replace(to_replace=np.Infinity, value=np.NaN).dropna()
name = data.columns.values.tolist()
names = list()
# data=data.sample(frac=0.2)
for i in name:
if i not in [
'stock_code', 'date', 'time', 'change_r', 'change_r_next',
'change_r_next16', 'change_r_next2', 'change_r_next3',
'change_r_next4', 'change_r_next48', '_id', 'classtype',
def deal_data(code):
data = KlineData.read_data(code=code,
start_date=dt.datetime(2015, 1, 1),
end_date=dt.datetime(2018, 9, 17),
kline='kline_day',
timemerge=True)
name = data.columns.values.tolist()
if len(data):
data = data.drop(['_id', 'classtype', 'market'], axis=1)
data = n_KDJ(data, 9, 3)
# t1 = time.clock()
# print(t1-t)
# t = time.clock()
data = n_MA(data)
# t1 = time.clock()
# print(t1 - t1)
# t = time.clock()
data = n_MACD(data, 12, 26, 9)
# t1 = time.clock()
# print(t1 - t)
# t = time.clock()
data = n_RSI(data, 6, 12, 24)
# t1 = time.clock()
# print(t1 - t)
# t = time.clock()
data = n_DMI(data, 14, 6)
# t1 = time.clock()
# print(t1 - t)
# t = time.clock()
data = n_BRAR(data, 26)
# t1 = time.clock()
# print(t1 - t)
# t = time.clock()
data = n_CR(data, 26, 10, 20, 40)
# t1 = time.clock()
# print(t1 - t)
# t = time.clock()
data = n_VR(data, 26)
# t1 = time.clock()
# print(t1 - t)
# t = time.clock()
data = n_WR(data, 10, 6)
# t1 = time.clock()
# print(t1 - t)
# t = time.clock()
data = n_CCI(data, 14)
# t1 = time.clock()
# print(t1 - t)
# t = time.clock()
data = n_BOLL(data, 20)
# t1 = time.clock()
# print(t1 - t)
# t = time.clock()
data = n_PSY(data, 12)
columns = data.columns.values.tolist()
data = data.replace(to_replace=np.Infinity, value=np.NaN)
fund_data = pd.DataFrame(
list(BaseModel('jq_fund_data').query({'code': code})))
jq_name = [
'capitalization', 'circulating_cap', 'circulating_market_cap',
'market_cap', 'pb_ratio', 'pcf_ratio', 'pe_ratio', 'pe_ratio_lyr',
'ps_ratio', 'turnover_ratio'
]
for i in jq_name:
data[i] = 0
# data['circulating_market_cap'] = 0
for i in columns:
if i not in name and 'change_r_next' not in i:
v = data[i].iloc[0]
if type(v).__name__ != 'bool_':
# print(i, v, type(v))
if data[i].min() == 0:
# pass
# print(i, data[i].loc[0])
data.loc[data[i] != 0,
i] = (data.loc[data[i] != 0, i] -
data.loc[data[i] != 0, i].min()) / (
data.loc[data[i] != 0, i].max() -
data.loc[data[i] != 0, i].min())
# print(i,data[i].loc[0])
else:
# print(i, v, type(v))
data[i] = (data[i] - data[i].min()) / (data[i].max() -
data[i].min())
# data['up'] = data['up'] / (data['up'] + data['down'])
# data['down'] = data['down'] / (data['up'] + data['down'])
data.loc[
(data.date >= dt.datetime(2015, 12, 31)) & (data.date < dt.datetime(2016, 3, 31)), jq_name] = \
fund_data[fund_data.day == '2015-12-31'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date >= dt.datetime(2016, 3, 31)) & (data.date <= dt.datetime(2016, 6, 30)), jq_name] = \
fund_data[fund_data.day == '2016-03-31'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date < dt.datetime(2016, 9, 30)) & (data.date >= dt.datetime(2016, 6, 30)), jq_name] = \
fund_data[fund_data.day == '2016-06-30'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date >= dt.datetime(2016, 9, 30)) & (data.date < dt.datetime(2016, 12, 30)), jq_name] = \
fund_data[fund_data.day == '2016-09-30'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date < dt.datetime(2017, 3, 31)) & (data.date >= dt.datetime(2016, 12, 30)), jq_name] = \
fund_data[fund_data.day == '2016-12-30'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date >= dt.datetime(2017, 3, 31)) & (data.date < dt.datetime(2017, 6, 30)), jq_name] = \
fund_data[fund_data.day == '2017-03-31'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date >= dt.datetime(2017, 6, 30)) & (data.date < dt.datetime(2017, 9, 29)), jq_name] = \
fund_data[fund_data.day == '2017-06-30'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date >= dt.datetime(2017, 9, 29)) & (data.date < dt.datetime(2017, 12, 29)), jq_name] = \
fund_data[fund_data.day == '2017-09-29'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date >= dt.datetime(2017, 12, 29)) & (data.date < dt.datetime(2018, 3, 30)), jq_name] = \
fund_data[fund_data.day == '2017-12-29'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date >= dt.datetime(2018, 3, 30)) & (data.date < dt.datetime(2018, 6, 29)), jq_name] = \
fund_data[fund_data.day == '2018-03-30'].loc[:,jq_name].iloc[0].tolist()
data.loc[
(data.date >= dt.datetime(2018, 6, 29)), jq_name] = \
fund_data[fund_data.day == '2018-06-29'].loc[:,jq_name].iloc[0].tolist()
name = ['open', 'close', 'high', 'low', 'amount', 'volume']
data['turnover_ratio'] = data.volume / (data.circulating_cap * 10000)
for i in name:
data[i] = (data[i] - data[i].min()) / (data[i].max() -
data[i].min())
data = data[::-1]
data = data.reset_index(drop=True)
# data['count'] = data.index.tolist()
curror = BaseModel('LaunchDate').query(sql={'stock_code': code})
data['count'] = 0
if curror.count():
launchdate = list(curror)[0]['date']
data['count'] = data.date.map(lambda x: (x - launchdate).days)
data = data[data.date >= dt.datetime(2016, 1, 1)]
BaseModel('features_kline_day').insert_batch(
data.to_dict(orient='records'))
from sklearn import tree
import pandas as pd
import datetime as dt
from Calf.models.base_model import BaseModel
from sklearn.metrics import classification_report
if __name__ == '__main__':
curror = BaseModel('novel_Feature').query(
sql=dict(stock_code=1,
date={
'$gte': dt.datetime(2018, 4, 15),
'$lte': dt.datetime(2018, 4, 23)
}))
print(curror.count())
data = pd.DataFrame(list(curror))
data['change_r_next'] = data.change_r_next.map(lambda x: 1 if x > 0 else 0)
data_test = data[data.date == dt.datetime(2018, 4, 23)]
data = data[data.date < dt.datetime(2018, 4, 23)]
y = data.loc[:, ['change_r_next']]
X = data.drop(['_id', 'change_r_next', 'date'], axis=1)
X_test = data_test.drop(['_id', 'change_r_next', 'date'], axis=1)
y_test = data_test.loc[:, ['change_r_next']]
clf = tree.DecisionTreeClassifier(class_weight=None,
criterion='gini',
max_depth=None,
max_leaf_nodes=None,
max_features=None,
