Skip to content

R
robo-dividend
Commit 89e819a5 authored by wenwen.tang's avatar wenwen.tang 😕
Browse files
Options

推荐系统

parent c418191a
Hide whitespace changes
Inline Side-by-side
Showing with 1 addition and 813 deletions
ai/EstimateMarketTrendV20.py deleted 100644 → 0
View file @ c418191a
import json
from typing import List
import numpy as np
from py_jftech import autowired, parse_date, prev_workday, format_date
from ai.config import LABEL_RANGE, LABEL_TAG
from ai.dao import robo_predict
from ai.dao.robo_datas import get_base_info, get_index_list, get_fund_list
from ai.data_access import DataAccess
from ai.model_trainer import ModelTrainer
from ai.noticer import send
from ai.training_data_builder import TrainingDataBuilder
from api import DataSync
# 截止日期
max_date = None
# max_date = '2024-03-20'
# max_date = '2024-01-11'
toForecast = True # False means test, True means forecast
syncData = True # 开启会同步数据库指数及基金数据
uploadData = True # 开启会上传预测结果
doReport = True # 开启会生成Excel报告
# 待预测指数
# PREDICT_LIST = [67, 121, 122, 123]
PREDICT_LIST = [67, 121, 122, 123, 147, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163,
164, 165, 166, 167, 168, 169, 170, 171, 174, 175, 177, 178]
eco = [65, 66, 74, 134]
index = [67, 68, 69, 70, 71, 72, 73, 75, 76, 77, 105, 106, 116, 117, 138, 139, 142, 143, 140, 141, 144, 145, 146]
fund = [121, 122, 123, 147, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165,
166, 167, 168, 169, 170, 171, 174, 175, 177, 178]
@autowired
def sync(syncs: List[DataSync] = None):
for s in syncs:
# if isinstance(s, (IndexSync, EcoSync)):
s.do_sync()
def report_prediction(label, predict_item, indexDict: dict):
prediction = label
predictionStr = LABEL_TAG.get(prediction)
content = f"""\n On day {forecastDay.strftime("%m/%d/%Y")}, the model predicts {predict_item} to be {predictionStr} in {str(numForecastDays)} business days. \n"""
print(content)
# 上传预测结果
key = [k for k, v in indexDict.items() if v == predict_item][0]
index_info = get_base_info(key)[0]
if uploadData:
if len(LABEL_RANGE) > 2:
data = {"rbd_id": key, "date": forecastDay, "predict": prediction}
robo_predict.insert(data)
else:
from ai.noticer import upload_predict
upload_predict(index_info['ticker'], forecastDay, predictionStr)
send(content)
return prediction
def judgement(id, type, predict):
from datetime import datetime
predict_term = 21
start = parse_date(max_date) if max_date else prev_workday(datetime.today())
navs = []
if type == 'INDEX':
navs = get_index_list(index_ids=id, min_date=start, limit=predict_term)
navs = [nav['rid_close'] for nav in navs]
elif type == 'FUND':
navs = get_fund_list(fund_ids=id, min_date=start, limit=predict_term)
navs = [nav['rfn_nav_cal'] for nav in navs]
if len(navs) == predict_term:
upper = True if navs[-1] >= navs[0] else False
result = {}
for k, v, in predict.items():
pred = True if v[0] > 0 else False
if upper == pred:
result[k] = True
else:
result[k] = False
j = {
'id': id,
'date': format_date(start),
'result': result
}
with open('predict.txt', 'a+') as file:
file.write(json.dumps(j))
file.write('\n')
########################################
if __name__ == '__main__':
if syncData:
sync()
# define some parameters
win1W = 5 # 1 week
win1M = 21 # 1 Month
win1Q = 63 # 1 Quarter
numForecastDays = 21 # business days, 21 business days means one month
theThreshold = 0.0
ids = set(PREDICT_LIST) | set(eco) | set(index) | set(fund)
infos = get_base_info(ids)
infos_type = {info['id']: info['type'] for info in infos}
indexDict = {info['id']: info['ticker'].replace(' Index', '').replace(' Equity', '').replace(' ', '_') for info in
infos}
###################
# Step 1: Prepare X and y (features and labels)
# 准备基础数据
data_access = DataAccess(index, eco, fund, max_date, indexDict)
indexData = data_access.get_index_datas()
ecoData = data_access.get_eco_datas()
fundData = data_access.get_fund_datas()
# 指数数据准备
vixData = data_access.get_vix(indexData)
indexOtherData = data_access.get_other_index(indexData)
# 经济指标数据准备
cpiData = data_access.get_cpi(ecoData)
FDTRData = data_access.get_fdtr(ecoData)
# 新增指标 NAPMPMI ：美國的ISM製造業指數 (Monthly)
NAPMPMIData = data_access.get_napmpmi(ecoData)
builder = TrainingDataBuilder(index, eco, fund, indexDict, toForecast, win1W, win1M, win1Q, numForecastDays,
theThreshold)
for pid in PREDICT_LIST:
print(f'{indexDict[pid]} start '.center(50, '='))
t_data = indexData if pid in index else fundData
X_train, X_test, y_train, y_test, scaledX_forecast, forecastDay = \
builder.build_train_test(pid, t_data, vixData, indexOtherData, cpiData, FDTRData, NAPMPMIData)
trainer = ModelTrainer(toForecast)
rf_model = trainer.train_random_forest(X_train, y_train, X_test, y_test)
gbt_model = trainer.train_GBT(X_train, y_train, X_test, y_test)
svc_model = trainer.train_SVC(X_train, y_train, X_test, y_test)
knn_model = trainer.train_nearest_neighbors(X_train, y_train, X_test, y_test)
ada_model = trainer.train_AdaBoost(X_train, y_train, X_test, y_test)
ensemble_model = trainer.ensemble_model(rf_model, gbt_model, svc_model,
knn_model, ada_model, X_train, y_train, X_test, y_test)
model_predict = {'forest': rf_model.predict(scaledX_forecast),
'gbt': gbt_model.predict(scaledX_forecast),
'svc': svc_model.predict(scaledX_forecast),
'knn': knn_model.predict(scaledX_forecast),
'adaboost': ada_model.predict(scaledX_forecast),
'ensemble': ensemble_model.predict(scaledX_forecast)}
print(f'预测结果：{model_predict}'.center(60, '+'))
judgement(pid, infos_type[pid], model_predict)
if toForecast:
if len(LABEL_RANGE) > 2:
average = round(np.mean(list(model_predict.values())))
report_prediction(average, indexDict[pid], indexDict)
else:
report_prediction(ensemble_model.predict(scaledX_forecast), indexDict[pid], indexDict)
if doReport:
if len(LABEL_RANGE) > 2:
from ai.reporter import do_reporter2
do_reporter2()
else:
from ai.reporter import do_reporter
do_reporter()
ai/config.py deleted 100644 → 0
View file @ c418191a
# 预测标签
from math import inf
LABEL_RANGE = {2: [0.05, inf], 1: [0.02, 0.05], 0: [-0.02, 0.02], -1: [-0.05, -0.02], -2: [-inf, -0.05]}
LABEL_TAG = {2: 'UPUP', 1: 'UP', 0: 'NEUTRAL', -1: 'DOWN', -2: 'DOWNDOWN'}
ai/dao/robo_datas.py deleted 100644 → 0
View file @ c418191a
from py_jftech import read, to_tuple, where
@read
def get_index_list(index_ids=None, min_date=None, max_date=None, limit=None):
limit_sql = f'limit {limit}' if limit else ''
sqls = []
if min_date:
sqls.append(f"rid_date >= '{min_date}'")
if max_date:
sqls.append(f"rid_date <= '{max_date}'")
return f'''
select * from robo_index_datas
{where(*sqls, rid_index_id=to_tuple(index_ids))} order by rid_index_id, rid_date {limit_sql}
'''
@read
def get_eco_list(eco_ids=None, min_date=None, max_date=None):
sqls = []
if min_date:
sqls.append(f"red_date >= '{min_date}'")
if max_date:
sqls.append(f"red_date <= '{max_date}'")
return f'''
select * from robo_eco_datas
{where(*sqls, red_eco_id=to_tuple(eco_ids))} order by red_eco_id, red_date
'''
@read
def get_fund_list(fund_ids=None, min_date=None, max_date=None, limit=None):
limit_sql = f'limit {limit}' if limit else ''
sqls = []
if min_date:
sqls.append(f"rfn_date >= '{min_date}'")
if max_date:
sqls.append(f"rfn_date <= '{max_date}'")
return f'''
select * from robo_fund_navs
{where(*sqls, rfn_fund_id=to_tuple(fund_ids))} order by rfn_fund_id, rfn_date {limit_sql}
'''
@read
def get_base_info(ids=None):
sqls = []
return f"""
SELECT rbd_id id,v_rbd_bloomberg_ticker ticker,v_rbd_type type, rbd_datas datas FROM `robo_base_datum`
{where(*sqls, rbd_id=to_tuple(ids))}
"""
ai/dao/robo_predict.py deleted 100644 → 0
View file @ c418191a
from py_jftech import read, write, format_date, to_tuple, where, mapper_columns
__COLUMNS__ = {
'rp_rbd_id': 'rbd_id',
'rp_date': 'date',
'rp_predict': 'predict',
'rp_remark': 'remark',
'rp_create_time': 'create_time'
}
@write
def insert(datas):
datas = mapper_columns(datas=datas, columns=__COLUMNS__)
return f'''
replace into robo_predict({','.join([x for x in datas.keys()])})
values ({','.join([f"'{x[1]}'" for x in datas.items()])})
'''
@read
def get_list(index_ids: object = None, min_date: object = None, max_date: object = None) -> object:
sqls = []
if min_date:
sqls.append(f"rp_date >= '{format_date(min_date)}'")
if max_date:
sqls.append(f"rp_date <= '{format_date(max_date)}'")
return f'''
select {','.join([f"{x[0]} as {x[1]}" for x in __COLUMNS__.items()])} from robo_predict
{where(*sqls, rid_index_id=to_tuple(index_ids))} order by rp_rbd_id, rp_date
'''
ai/dao/robo_predict.sql deleted 100644 → 0
View file @ c418191a
CREATE TABLE IF NOT EXISTS robo_predict (
`rp_rbd_id` bigint(20) NOT NULL,
`rp_date` datetime NOT NULL,
`rp_predict` int(11) NOT NULL,
`rp_remark` json NULL,
`rp_create_time` datetime NOT NULL DEFAULT CURRENT_TIMESTAMP,
`rp_update_time` datetime NULL DEFAULT NULL ON UPDATE CURRENT_TIMESTAMP,
INDEX `rp_rbd_id`(`rp_rbd_id`, `rp_date`) USING BTREE
) ENGINE = InnoDB CHARACTER SET = utf8mb4 COLLATE = utf8mb4_general_ci ROW_FORMAT = Dynamic;
\ No newline at end of file
ai/data_access.py deleted 100644 → 0
View file @ c418191a
from abc import ABC
import pandas as pd
from ai.dao.robo_datas import get_eco_list, get_fund_list, get_index_list
class DataAccess(ABC):
def __init__(self, index, eco, fund, max_date, indexDict) -> None:
super().__init__()
self._index = index
self._eco = eco
self._fund = fund
self._max_date = max_date
self._indexDict = indexDict
def get_index_datas(self):
indexData = pd.DataFrame(
get_index_list(index_ids=self._index, max_date=self._max_date))
# todo erp 没有数据 "rid_erp",
indexData = indexData[
["rid_index_id", "rid_date", "rid_high", "rid_open", "rid_low", "rid_close", "rid_pe", "rid_pb",
"rid_volume", "rid_frdpe", "rid_frdpes", "rid_pc"]]
# 数据替换和截取以后用105 和106，（都从2022.1.3开始）然后2022年以前的部分：105前边接76的数据 106前边接77的数据
condition1 = ((indexData['rid_index_id'] == 76) & (indexData['rid_date'] >= '2022-01-03'))
condition2 = ((indexData['rid_index_id'] == 77) & (indexData['rid_date'] >= '2022-01-03'))
condition3 = ((indexData['rid_index_id'] == 105) & (indexData['rid_date'] < '2022-01-03'))
condition4 = ((indexData['rid_index_id'] == 106) & (indexData['rid_date'] < '2022-01-03'))
indexData.drop(indexData[condition1 | condition2 | condition3 | condition4].index, inplace=True)
indexData.loc[indexData['rid_index_id'] == 76, 'rid_index_id'] = 105
indexData.loc[indexData['rid_index_id'] == 77, 'rid_index_id'] = 106
indexData.rename(columns={"rid_date": 'date'}, inplace=True) # please use 'date'
indexData["rid_index_id"] = indexData["rid_index_id"].map(self._indexDict)
return indexData
def get_eco_datas(self):
ecoData = pd.DataFrame(
get_eco_list(eco_ids=self._eco, max_date=self._max_date))
ecoData = ecoData[["red_eco_id", "red_date", "red_indicator"]]
ecoData.rename(columns={"red_date": 'date'}, inplace=True) # please use 'date'
ecoData["red_eco_id"] = ecoData["red_eco_id"].map(self._indexDict)
return ecoData
def get_fund_datas(self):
fundData = pd.DataFrame(
get_fund_list(fund_ids=self._fund, max_date=self._max_date))
fundData = fundData[["rfn_fund_id", "rfn_date", "rfn_nav_cal"]]
fundData.rename(columns={"rfn_date": 'date'}, inplace=True) # please use 'date'
fundData["rfn_fund_id"] = fundData["rfn_fund_id"].map(self._indexDict)
return fundData
def get_vix(self, indexData):
# VIX：芝加哥期权交易所SPX波动率指
vixData = indexData[indexData['rid_index_id'] == "VIX"].copy()
vixData = vixData[
["date", "rid_high", "rid_open", "rid_low", "rid_close", "rid_pc", "rid_pb", "rid_pe", "rid_frdpe",
"rid_frdpes"]]
vixData.rename(
columns={"rid_high": 'vix_high', 'rid_open': 'vix_open', "rid_low": 'vix_low', "rid_volume": 'vix_volume',
"rid_close": 'vix_close', "rid_pc": 'vix_pc', "rid_pb": 'vix_pb', "rid_pe": 'vix_pe',
"rid_frdpe": 'vix_frdpe', "rid_frdpes": 'vix_frdpes'},
inplace=True)
vixData.set_index('date', inplace=True)
vixData.index = pd.to_datetime(vixData.index)
vixData.dropna(axis=1, inplace=True)
return vixData
def get_other_index(self, indexData):
other_index = ["USGG10YR", "USGG2YR", "CCMP", "TSFR1M", "TSFR12M", "COI_TOTL", "LEI_TOTL", "MID",
"OE4EKLAC", "OEA5KLAC", "OECNKLAC", "OEJPKLAC", "OEOTGTAC", "OEUSKLAC", "USRINDEX", "SPX"]
cols = ['date', 'rid_close', 'rid_pe', 'rid_pb', 'rid_volume', 'rid_frdpe', 'rid_frdpes', 'rid_pc']
indexOtherData = pd.DataFrame()
idxs = [self._indexDict[i] for i in self._index]
for idx in other_index:
if idx in idxs:
idx_data = indexData[indexData['rid_index_id'] == idx].copy()
idx_data = idx_data[cols]
idx_data.rename(
columns={"rid_close": f'{idx}_close', 'rid_pe': f'{idx}_pe', 'rid_pb': f'{idx}_pb',
'rid_volume': f'{idx}_volume', 'rid_frdpe': f'{idx}_frdpe', 'rid_frdpes': f'{idx}_frdpes',
'rid_pc': f'{idx}_pc'},
inplace=True)
idx_data.set_index('date', inplace=True)
idx_data.index = pd.to_datetime(idx_data.index)
if indexOtherData.size > 0:
indexOtherData = pd.merge(indexOtherData, idx_data, how='outer', on='date')
else:
indexOtherData = idx_data
indexOtherData.ffill(inplace=True)
indexOtherData.bfill(inplace=True)
indexOtherData = indexOtherData.dropna(axis=1)
return indexOtherData
def get_cpi(self, ecoData):
# CPI_YOY:美国城镇消费物价指数同比未经季 CPURNSA:美国消费者物价指数未经季调
cpiData = ecoData[(ecoData['red_eco_id'] == "CPI_YOY") | (ecoData['red_eco_id'] == "CPURNSA")].copy()
cpiData = cpiData.pivot(index='date', columns='red_eco_id', values='red_indicator')
cpiData['CPI_MOM'] = (cpiData['CPURNSA'] / cpiData['CPURNSA'].shift(
1) - 1.0) * 100 * 12 # Annualized Percentage
cpiData['CPI_MOM_Diff'] = cpiData['CPURNSA'] - cpiData['CPURNSA'].shift(1)
cpiData.index = pd.to_datetime(cpiData.index)
return cpiData
def get_fdtr(self, ecoData):
# FDTR 美国联邦基金目标利率
FDTRData = ecoData[ecoData['red_eco_id'] == "FDTR"].copy()
del (FDTRData['red_eco_id'])
FDTRData.rename(columns={"red_indicator": 'FDTR'}, inplace=True)
FDTRData.set_index('date', inplace=True)
FDTRData.index = pd.to_datetime(FDTRData.index)
return FDTRData
def get_napmpmi(self, ecoData):
# 新增指标 NAPMPMI ：美國的ISM製造業指數 (Monthly)
NAPMPMIData = ecoData[ecoData['red_eco_id'] == "NAPMPMI"].copy()
del (NAPMPMIData['red_eco_id'])
NAPMPMIData.rename(columns={"red_indicator": 'NAPMPMI'}, inplace=True)
NAPMPMIData.set_index('date', inplace=True)
NAPMPMIData.index = pd.to_datetime(NAPMPMIData.index)
return NAPMPMIData
ai/model_trainer.py deleted 100644 → 0
View file @ c418191a
from abc import ABC
import matplotlib.pyplot as plt
from lightgbm import LGBMClassifier
from sklearn import svm
from sklearn.ensemble import AdaBoostClassifier
from sklearn.ensemble import RandomForestClassifier, VotingClassifier
from sklearn.metrics import classification_report, confusion_matrix, ConfusionMatrixDisplay, accuracy_score
from sklearn.neighbors import KNeighborsClassifier
from ai.config import LABEL_RANGE
class ModelTrainer(ABC):
"""
模型训练类
"""
def __init__(self, toForecast) -> None:
super().__init__()
self._toForecast = toForecast
###################
# Step 3: Train the model
def test_model(self, strMethod, classifier, X_test, y_test):
print(strMethod + " ====== test results ======")
y_pred = classifier.predict(X_test)
labels = list(LABEL_RANGE.keys())[::-1]
result0 = confusion_matrix(y_test, y_pred, labels=labels)
print(strMethod + " Confusion Matrix:")
print(result0)
result1 = classification_report(y_test, y_pred, zero_division=1.0)
print(strMethod + " Classification Report:")
print(result1)
result2 = accuracy_score(y_test, y_pred)
print(strMethod + " Accuracy:", result2)
cm_display = ConfusionMatrixDisplay(confusion_matrix=result0, display_labels=labels)
cm_display.plot()
plt.title(strMethod + ' Accuracy: ' + f'{result2:.0%}')
plt.show()
def train_random_forest(self, X_train, y_train, X_test, y_test):
classifier = RandomForestClassifier()
classifier.fit(X_train, y_train)
if not self._toForecast:
self.test_model('Random Forest', classifier, X_test, y_test)
return classifier
def train_GBT(self, X_train, y_train, X_test, y_test):
# Gradient Boosted Tree
classifierGBT = LGBMClassifier()
classifierGBT.fit(X_train, y_train)
if not self._toForecast:
self.test_model('Gradient Boosted Tree', classifierGBT, X_test, y_test)
return classifierGBT
def train_SVC(self, X_train, y_train, X_test, y_test):
# Support Vector Machines
classifierSVC = svm.SVC()
classifierSVC.fit(X_train, y_train)
if not self._toForecast:
self.test_model('Support Vector Machines', classifierSVC, X_test, y_test)
return classifierSVC
def train_nearest_neighbors(self, X_train, y_train, X_test, y_test):
classifier = KNeighborsClassifier()
classifier.fit(X_train, y_train)
if not self._toForecast:
self.test_model('K-Nearest Neighbors', classifier, X_test, y_test)
return classifier
def train_AdaBoost(self, X_train, y_train, X_test, y_test):
classifier = AdaBoostClassifier()
classifier.fit(X_train, y_train)
if not self._toForecast:
self.test_model('AdaBoost', classifier, X_test, y_test)
return classifier
def ensemble_model(self, rf_model, gbt_model, svc_model, knn_model,
ada_model, X_train, y_train, X_test, y_test):
# Create a dictionary of our models
estimators = [('rf', rf_model), ('gbt', gbt_model), ('svc', svc_model),
('knn', knn_model), ('AdaBoost', ada_model)]
# Create our voting classifier, inputting our models
ensemble = VotingClassifier(estimators, voting='hard')
# fit model to training data
ensemble.fit(X_train, y_train)
if not self._toForecast:
self.test_model('Ensemble Model', ensemble, X_test, y_test)
return ensemble
ai/noticer.py deleted 100644 → 0
View file @ c418191a
from datetime import datetime
import requests
from py_jftech import sendmail, format_date
# 预测发送邮箱
email = ['wenwen.tang@thizgroup.com']
jrp_domain = 'https://jrp.jfquant.com/api/v1.0'
# jrp_domain = 'http://localhost:7090/jrp'
def send(content):
receives = email
subject = '预测_{today}'.format(today=format_date(datetime.today()))
sendmail(receives=receives, copies=[], attach_paths=[], subject=subject, content=content)
def upload_predict(ticker, predictDate, predict):
predict_data = {
"aiPredict": {
"predictDate": format_date(predictDate),
"predict": 1 if predict == 'UP' else -1
},
"bloombergTicker": ticker
}
headers = {"X-AUTH-token": "rt7297LwQvyAYTke2iD8Vg"}
response = requests.post(url=f'{jrp_domain}/ai/predict', json=predict_data, headers=headers)
if response.status_code != 200:
print(response.text)
print("上传ai预测结果失败，请重试")
ai/predict.txt deleted 100644 → 0
View file @ c418191a
{"id": 67, "date": "2023-10-06", "result": {"forest": false, "gbt": true, "svc": true, "ensemble": true}}
{"id": 121, "date": "2023-10-06", "result": {"forest": true, "gbt": true, "svc": false, "ensemble": true}}
{"id": 122, "date": "2023-10-06", "result": {"forest": true, "gbt": true, "svc": false, "ensemble": true}}
{"id": 123, "date": "2023-10-06", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 67, "date": "2023-10-13", "result": {"forest": false, "gbt": true, "svc": true, "ensemble": true}}
{"id": 121, "date": "2023-10-13", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 122, "date": "2023-10-13", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 123, "date": "2023-10-13", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 67, "date": "2023-10-20", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 121, "date": "2023-10-20", "result": {"forest": false, "gbt": true, "svc": false, "ensemble": true}}
{"id": 122, "date": "2023-10-20", "result": {"forest": true, "gbt": true, "svc": false, "ensemble": true}}
{"id": 123, "date": "2023-10-20", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 67, "date": "2023-10-27", "result": {"forest": false, "gbt": true, "svc": true, "ensemble": true}}
{"id": 121, "date": "2023-10-27", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 122, "date": "2023-10-27", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 123, "date": "2023-10-27", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 67, "date": "2023-11-03", "result": {"forest": true, "gbt": true, "svc": true, "ensemble": true}}
{"id": 121, "date": "2023-11-03", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 122, "date": "2023-11-03", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 123, "date": "2023-11-03", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 67, "date": "2023-11-17", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 121, "date": "2023-11-17", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 122, "date": "2023-11-17", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 123, "date": "2023-11-17", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 67, "date": "2023-11-24", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 121, "date": "2023-11-24", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 122, "date": "2023-11-24", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 123, "date": "2023-11-24", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 67, "date": "2023-12-01", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 121, "date": "2023-12-01", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 122, "date": "2023-12-01", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 123, "date": "2023-12-01", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 67, "date": "2023-12-08", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 121, "date": "2023-12-08", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 122, "date": "2023-12-08", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 123, "date": "2023-12-08", "result": {"forest": false, "gbt": false, "svc": false, "ensemble": false}}
{"id": 67, "date": "2023-12-15", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 121, "date": "2023-12-15", "result": {"forest": true, "gbt": true, "svc": true, "ensemble": true}}
{"id": 122, "date": "2023-12-15", "result": {"forest": false, "gbt": false, "svc": true, "ensemble": false}}
{"id": 123, "date": "2023-12-15", "result": {"forest": true, "gbt": true, "svc": true, "ensemble": true}}
ai/reporter.py deleted 100644 → 0
View file @ c418191a
import datetime
import json
import pandas as pd
import requests
from ai.config import LABEL_TAG
from ai.dao import robo_predict
from ai.dao.robo_datas import get_base_info
symbols = ['ACWI', 'EWJ', 'MCHI', 'EEM', 'BKF', 'INDA', 'AAXJ', 'VGK', 'QQQ', 'SPY', 'SPX', 'IWN',
'IUSG', 'IWD', 'DON', 'GDX', 'TOLZ', 'XLU', 'XBI', 'ESGD', 'IGE', 'EMLC', 'IGAA',
'LQD', 'HYG', 'SHY', 'IEI', 'IEF', 'GLD', 'IYR', 'UUP', 'CEW', 'TLT']
def do_reporter(start='2023-10-01', end=datetime.date.today()):
url = f"https://jrp.jfquant.com/api/v1.0/ai/predict?startTime={start}&endTime={end.strftime('%Y-%m-%d')}"
resp = requests.get(url)
datas = []
symbol_index_dict = {symbol: index for index, symbol in enumerate(symbols)}
for value in resp.json()['body'].values():
for item in value:
data = {
'Forcast On Date': item['aiPredict']['predictDate'],
'Ticker': item['bloombergTicker'].replace(' Index', '').replace(' Equity', ''),
'In 21 business days': 'UP' if item['aiPredict']['predict'] == 1 else 'DOWN',
'Ticker Name': item['indexName'],
}
datas.append(data)
sorted_data = sorted(datas, key=lambda x: symbol_index_dict[x['Ticker'].split(' ')[0]])
print(json.dumps(sorted_data, ensure_ascii=False))
pf = pd.DataFrame(sorted_data)
pf.to_excel("Forcast_Report.xlsx", index=False)
def do_reporter2():
datas = []
index_info = get_base_info()
info = {x['id']: x for x in index_info}
symbol_index_dict = {symbol: index for index, symbol in enumerate(symbols)}
records = robo_predict.get_list()
for item in records:
data = {
'Forcast On Date': item['date'],
'Ticker': info.get(item['rbd_id'])['ticker'].replace(' Index', '').replace(' Equity', ''),
'In 21 business days': LABEL_TAG.get(item['predict']),
'Ticker Name': json.loads(info.get(item['rbd_id'])['datas'])['chineseName'],
}
datas.append(data)
sorted_data = sorted(datas, key=lambda x: symbol_index_dict[x['Ticker'].split(' ')[0]])
pf = pd.DataFrame(sorted_data)
pf.to_excel("Forcast_Report.xlsx", index=False)
if __name__ == '__main__':
do_reporter2()
ai/training_data_builder.py deleted 100644 → 0
View file @ c418191a
from abc import ABC
import numpy as np
import pandas as pd
from finta import TA
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import MinMaxScaler
from ai.config import LABEL_RANGE
def imp():
print(TA)
class TrainingDataBuilder(ABC):
def __init__(self, index, eco, fund, indexDict, toForecast, win1W, win1M, win1Q, numForecastDays,
theThreshold) -> None:
super().__init__()
self._index = index
self._eco = eco
self._fund = fund
self._indexDict = indexDict
self._toForecast = toForecast
self._win1W = win1W # 1 week
self._win1M = win1M # 1 Month
self._win1Q = win1Q # 1 Quarter
self._numForecastDays = numForecastDays # business days, 21 business days means one month
self._theThreshold = theThreshold
# List of symbols for technical indicators
# INDICATORS = ['RSI', 'MACD', 'STOCH','ADL', 'ATR', 'MOM', 'MFI', 'ROC', 'OBV', 'CCI', 'EMV', 'VORTEX']
# Note that '14 period MFI' and '14 period EMV' is not available for forecast
self.INDICATORS = ['RSI', 'MACD', 'STOCH', 'ADL', 'ATR', 'MOM', 'ROC', 'OBV', 'CCI', 'VORTEX']
self.FUND_INDICATORS = []
def get_indicator_data(self, data, pid):
"""
Function that uses the finta API to calculate technical indicators used as the features
"""
def indicator_calcu(data, indicators):
"""
指数和基金不同，基金只有收盘价，生成指标会变少
@param data:
@param indicators:
@return:
"""
for indicator in indicators:
ind_data = eval('TA.' + indicator + '(data)')
if not isinstance(ind_data, pd.DataFrame):
ind_data = ind_data.to_frame()
data = data.merge(ind_data, left_index=True, right_index=True)
return data
if pid in self._index:
data = indicator_calcu(data, self.INDICATORS)
# Instead of using the actual volume value (which changes over time), we normalize it with a moving volume average
data['normVol'] = data['volume'] / data['volume'].ewm(5).mean()
# get relative values
data['relativeOpen'] = data['open'] / data['close'].shift(1)
data['relativeHigh'] = data['high'] / data['close'].shift(1)
data['relativeLow'] = data['low'] / data['close'].shift(1)
# Remove columns that won't be used as features
# data['close'] are still needed and will be deleted later
data.drop(['open', 'high', 'low', 'volume'], axis=1, inplace=True)
elif pid in self._fund:
indicator_calcu(data, self.FUND_INDICATORS)
# Also calculate moving averages for features
data['ema50'] = data['close'] / data['close'].ewm(50).mean()
data['ema21'] = data['close'] / data['close'].ewm(21).mean()
data['ema15'] = data['close'] / data['close'].ewm(15).mean()
data['ema5'] = data['close'] / data['close'].ewm(5).mean()
data['relativeClose'] = data['close'] / data['close'].shift(1)
return data
def build_predict_data(self, indexData, pid):
def map_to_label(ret):
for label, (lower, upper) in LABEL_RANGE.items():
if float(lower) <= ret < float(upper):
return label
"""
@param pid: 需要预测的指数或基金id
@return:
"""
if pid in self._index:
###### get individual data from raw data
predictData = indexData[indexData['rid_index_id'] == self._indexDict[pid]].copy()
del (predictData['rid_index_id'])
###### Additional preparing SPX Data
# finta expects properly formated ohlc DataFrame, with column names in lowercase:
# ["open", "high", "low", close"] and ["volume"] for indicators that expect ohlcv input.
predictData.rename(
columns={"rid_high": 'high', 'rid_open': 'open', "rid_low": 'low', "rid_close": 'close',
'rid_volume': 'volume',
"rid_pe": f"{self._indexDict[pid]}_pe", "rid_pb": f"{self._indexDict[pid]}_pb"},
inplace=True)
elif pid in self._fund:
predictData = indexData[indexData['rfn_fund_id'] == self._indexDict[pid]].copy()
del (predictData['rfn_fund_id'])
predictData.rename(columns={"rfn_nav_cal": 'close'}, inplace=True)
predictData.set_index('date', inplace=True)
predictData.index = pd.to_datetime(predictData.index)
predictData.sort_index(inplace=True)
predictData.reset_index(inplace=True)
# Calculate the indicator data
predictData = self.get_indicator_data(predictData, pid)
# Calculate Historical Return and Volatility
predictData['R1W'] = np.log(predictData['close'] / predictData['close'].shift(self._win1W))
predictData['R1M'] = np.log(predictData['close'] / predictData['close'].shift(self._win1M))
predictData['R1Q'] = np.log(predictData['close'] / predictData['close'].shift(self._win1Q))
price_list = predictData['close']
rollist = price_list.rolling(self._win1W)
predictData['Vol_1W'] = rollist.std(ddof=0)
rollist = price_list.rolling(self._win1M)
predictData['Vol_1M'] = rollist.std(ddof=0)
rollist = price_list.rolling(self._win1Q)
predictData['Vol_1Q'] = rollist.std(ddof=0)
# The following uses future info for the y label, to be deleted later
predictData['futureR'] = np.log(predictData['close'].shift(-self._numForecastDays) / predictData['close'])
# predictData = predictData[predictData['futureR'].notna()]
predictData['yLabel'] = predictData['futureR'].apply(lambda r: map_to_label(r))
del (predictData['close'])
return predictData
def build_train_test(self, pid, indexData, vixData, indexOtherData, cpiData, FDTRData, NAPMPMIData):
###### Merge Data to one table
predictData = self.build_predict_data(indexData, pid)
forecastDay = None
if (self._toForecast):
forecastDay = predictData['date'].iloc[-1]
DataAll = pd.merge(predictData, vixData, how='outer', on='date')
DataAll = pd.merge(DataAll, indexOtherData, how='outer', on='date')
DataAll = pd.merge(DataAll, cpiData, how='outer', on='date')
DataAll = pd.merge(DataAll, FDTRData, how='outer', on='date')
DataAll = pd.merge(DataAll, NAPMPMIData, how='outer', on='date')
DataAll.set_index('date', inplace=True)
DataAll.sort_index(inplace=True)
DataAll.reset_index(inplace=True)
###### fill eco data
for col in ['CPI_YOY', 'CPURNSA', 'CPI_MOM', 'CPI_MOM_Diff']:
DataAll[col].bfill(inplace=True)
for col in ['FDTR']:
DataAll[col].ffill(inplace=True)
# 新增指数NAPMPMI ：美國的ISM製造業指數 (Monthly)
for col in ['NAPMPMI']:
DataAll[col].bfill(inplace=True)
DataAll[col].ffill(inplace=True)
for col in DataAll.columns:
if col not in ['CPI_YOY', 'CPURNSA', 'CPI_MOM', 'CPI_MOM_Diff', 'futureR', 'yLabel']:
DataAll[col].ffill(inplace=True)
if (self._toForecast):
# 处理CPI_YOY:美国城镇消费物价指数同比未经季 CPURNSA:美国消费者物价指数未经季调
DataAllCopy = DataAll.copy()
for col in ['CPI_YOY', 'CPURNSA']:
DataAllCopy[col].ffill(inplace=True)
for col in ['CPI_MOM', 'CPI_MOM_Diff']:
DataAllCopy[col] = DataAllCopy[col].fillna(0)
DataAllCopy.drop(['futureR', 'yLabel'], axis=1, inplace=True)
forecastDayIndex = DataAllCopy.index[DataAllCopy['date'] == forecastDay]
forecastData = DataAllCopy.iloc[forecastDayIndex.to_list(), 1:]
forecastData.dropna(inplace=True, axis=1)
X_forecast = forecastData.to_numpy()
del DataAllCopy
###### clean NaN
DataAll.dropna(inplace=True)
DataAll.reset_index(inplace=True, drop=True)
###### get X and y
y = DataAll['yLabel'].to_numpy(copy=True)
# delete future information
DataAll.drop(['futureR', 'yLabel'], axis=1, inplace=True)
X = DataAll.iloc[:, 1:].values
###################
# scale data
labels = list(LABEL_RANGE.keys())
scaler = MinMaxScaler(feature_range=(labels[-1], labels[0]))
# scaledX = scaler.fit_transform(X)
DataScaler = scaler.fit(X)
scaledX = DataScaler.transform(X)
scaledX_forecast = None
if (self._toForecast):
scaledX_forecast = DataScaler.transform(X_forecast)
X_train = scaledX
y_train = y
X_test = []
y_test = []
else:
# Step 2: Split data into train set and test set
X_train, X_test, y_train, y_test = train_test_split(scaledX, y, test_size=0.02, shuffle=False)
# To avoid data leak, test set should start from numForecastDays later
X_test = X_test[self._numForecastDays:]
y_test = y_test[self._numForecastDays:]
return X_train, X_test, y_train, y_test, scaledX_forecast, forecastDay
requirements.txt
View file @ 89e819a5
......@@ -22,13 +22,4 @@ six==1.16.0
urllib3==1.26.12
fastapi==0.100.0
uvicorn==0.23.1
apscheduler==3.10.1
sklearn
finta
keras
tensorflow
matplotlib
lightgbm
\ No newline at end of file
apscheduler==3.10.1
\ No newline at end of file
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment