掘金社区
用pytorch 重新编写《策略研究》中“机器学习（股票）”的例子Pinned highlighted
zaomuwu
发表在策略研究 2020-01-25 22:13:59
coding=utf-8
from future import print_function, absolute_import, unicode_literals
from datetime import datetime
import numpy as np
from gm.api import *
import sys
import sklearn.datasets
import torch
import torch.nn as nn
import torch.nn.functional as F
try:
from sklearn import svm
from torch.autograd import Variable
except:
print('请安装scikit-learn、pytorch库和带mkl的numpy')
sys.exit(-1)
'''
本策略选取了七个特征变量组成了滑动窗口长度为15天的训练集,随后训练了一个二分类(上涨/下跌)的支持向量机模型.
若没有仓位则在每个星期一的时候输入标的股票近15个交易日的特征变量进行预测,并在预测结果为上涨的时候购买标的.
若已经持有仓位则在盈利大于10%的时候止盈,在星期五损失大于2%的时候止损.
特征变量为:1.收盘价/均值2.现量/均量3.最高价/均价4.最低价/均价5.现量6.区间收益率7.区间标准差
训练数据为:SHSE.600009上海机场,时间从2016-04-01到2017-07-30
回测时间为:2017-08-01 09:00:00到2017-09-05 09:00:00
'''
def init(context):
# 订阅上海机场的分钟bar行情
context.symbol = 'SHSE.600009'
subscribe(symbols=context.symbol, frequency='60s')
start_date = '2016-03-01' # SVM训练起始时间
end_date = '2017-07-30' # SVM训练终止时间
# 用于记录工作日
# 获取目标股票的daily历史行情
recent_data = history(symbol=context.symbol, frequency='1d', start_time=start_date, end_time=end_date, fill_missing='Last',
df=True)
days_value = recent_data['bob'].values
days_close = recent_data['close'].values
days = []
# 获取行情日期列表
print('准备数据训练SVM')
for i in range(len(days_value)):
days.append(str(days_value[i])[0:10])
x_all = []
y_all = []
for index in range(15, (len(days) - 5)):
# 计算三星期共15个交易日相关数据
start_day = days[index - 15]
end_day = days[index]
data = history(symbol=context.symbol, frequency='1d', start_time=start_day, end_time=end_day, fill_missing='Last',
df=True)
close = data['close'].values
max_x = data['high'].values
min_n = data['low'].values
amount = data['amount'].values
volume = []
for i in range(len(close)):
volume_temp = amount[i] / close[i]
volume.append(volume_temp)
close_mean = close[-1] / np.mean(close) # 收盘价/均值
volume_mean = volume[-1] / np.mean(volume) # 现量/均量
max_mean = max_x[-1] / np.mean(max_x) # 最高价/均价
min_mean = min_n[-1] / np.mean(min_n) # 最低价/均价
vol = volume[-1] # 现量
return_now = close[-1] / close[0] # 区间收益率
std = np.std(np.array(close), axis=0) # 区间标准差
# 将计算出的指标添加到训练集X
# features用于存放因子
features = [close_mean, volume_mean, max_mean, min_mean, vol, return_now, std]
x_all.append(features)
# 准备算法需要用到的数据
for i in range(len(days_close) - 20):
if days_close[i + 20] > days_close[i + 15]:
label = 1
else:
label = 0
y_all.append(label)
#print('y='np.shape(y_all))
x_train = x_all[: -1]
y_train = y_all[: -1]
# 训练SVM
#context.clf = svm.SVC(C=1.0, kernel='rbf', degree=3, gamma='auto', coef0=0.0, shrinking=True, probability=False,
# tol=0.001, cache_size=400, verbose=False, max_iter=-1,
# decision_function_shape='ovr', random_state=None)
#context.clf.fit(x_train, y_train)
#X = torch.from_numpy(x_train).type(torch.FloatTensor)
X=np.array(x_train)
y=np.array(y_train)
print('x=',np.shape(X))
X = torch.from_numpy(X).type(torch.FloatTensor)
context.X=Variable(X)
y = torch.from_numpy(y).type(torch.LongTensor)
context.y=Variable(y)
print(context.y)
#our class must extend nn.Module
net1 = torch.nn.Sequential(
torch.nn.Linear(7, 5),
torch.nn.Tanh(),
torch.nn.Linear(5, 2)
)
optimizer = torch.optim.SGD(net1.parameters(), lr=0.01,momentum=0.9)
loss_func = torch.nn.CrossEntropyLoss()
for t in range(100):
prediction =net1(context.X)
print(prediction,'prediction',context.y,'context.y')
#prediction =prediction.unsqeeze(1)
loss = loss_func(prediction, context.y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
print(loss,'loss',t)
torch.save(net1, 'net.pkl') # save entire net
torch.save(net1.state_dict(), 'net_params.pkl')
context.net2 = torch.load('net.pkl')
#prediction = net2(x)
def on_bar(context, bars):
bar = bars[0]
# 获取当前年月日
today = bar.bob.strftime('%Y-%m-%d')
last_day = get_previous_trading_date(exchange='SHSE', date=today)
# 获取数据并计算相应的因子
# 于星期一的09:31:00进行操作
# 当前bar的工作日
weekday = datetime.strptime(today, '%Y-%m-%d').isoweekday()
# 获取模型相关的数据
# 获取持仓
position = context.account().position(symbol=context.symbol, side=PositionSide_Long)
# 如果bar是新的星期一且没有仓位则开始预测
if not position and weekday == 1:
# 获取预测用的历史数据
data = history_n(symbol=context.symbol, frequency='1d', end_time=last_day, count=15,
fill_missing='Last', adjust=ADJUST_PREV, df=True)
close = data['close'].values
train_max_x = data['high'].values
train_min_n = data['low'].values
train_amount = data['amount'].values
volume = []
for i in range(len(close)):
volume_temp = train_amount[i] / close[i]
volume.append(volume_temp)
close_mean = close[-1] / np.mean(close)
volume_mean = volume[-1] / np.mean(volume)
max_mean = train_max_x[-1] / np.mean(train_max_x)
min_mean = train_min_n[-1] / np.mean(train_min_n)
vol = volume[-1]
return_now = close[-1] / close[0]
std = np.std(np.array(close), axis=0)
# 得到本次输入模型的因子
features = [close_mean, volume_mean, max_mean, min_mean, vol, return_now, std]
features = np.array(features).reshape(1, -1)
features=torch.from_numpy(features).type(torch.FloatTensor)
features=Variable(features)
prediction = context.net2(features)
prediction =F.softmax(prediction)
prediction = torch.max(prediction, 1)[1]
print(' prediction',prediction)
# 若预测值为上涨则开仓
if prediction == 1:
# 获取昨收盘价
context.price = close[-1]
# 把浦发银行的仓位调至95%
order_target_percent(symbol=context.symbol, percent=0.95, order_type=OrderType_Market,
position_side=PositionSide_Long)
print(context.symbol, '以市价单开多仓到仓位0.95')
# 当涨幅大于10%,平掉所有仓位止盈
elif position and bar.close / context.price >= 1.10:
order_close_all()
print(context.symbol, '以市价单全平多仓止盈')
# 当时间为周五并且跌幅大于2%时,平掉所有仓位止损
elif position and bar.close / context.price < 1.02 and weekday == 5:
order_close_all()
print(context.symbol, '以市价单全平多仓止损')
if name == 'main':
'''
strategy_id策略ID,由系统生成
filename文件名,请与本文件名保持一致
mode实时模式:MODE_LIVE回测模式:MODE_BACKTEST
token绑定计算机的ID,可在系统设置-密钥管理中生成
backtest_start_time回测开始时间
backtest_end_time回测结束时间
backtest_adjust股票复权方式不复权:ADJUST_NONE前复权:ADJUST_PREV后复权:ADJUST_POST
backtest_initial_cash回测初始资金
backtest_commission_ratio回测佣金比例
backtest_slippage_ratio回测滑点比例
'''
run(strategy_id='strategy_id',
filename='main.py',
mode=MODE_BACKTEST,
token='token_id',
backtest_start_time='2017-08-01 09:00:00',
backtest_end_time='2017-09-05 09:00:00',
backtest_adjust=ADJUST_PREV,
backtest_initial_cash=10000000,
backtest_commission_ratio=0.0001,
backtest_slippage_ratio=0.0001)
工商网监