def cal_signal_(df0, df1, df2, strat_time, end_time, cs0, sxf=2/1000, slip=2/1000):
'''
df1:所有的基础数据：时间，名称，ohlcv—vc，是否可买入，统一平仓日，统一换仓日
df2：对应的仓位利润：时间，名称，是否选中，是否持仓，持仓价格，持仓股数，
持仓金额，持仓利润率，持仓利润，单日涨跌幅，单日浮动盈亏，手续费，滑点金额
'''
# 配置参数:止盈，ma_len（atr），atr_倍数，交易次数
stop_win_n, ma_len, atrn,sw_atrn,da_ma_n, trade_inner_nums = cs0
# 配置临时变量
inner_trade_nums = trade_inner_nums
open_pos_size = 1
max_log_high = np.nan
min_log_low = np.nan
for i in range(10, df0.shape[0]): # 类似于vnpy:on_bar
# 换仓日
if df0[i][-1]:
# 换仓操作
elif
# # 交易时间判断：日：1-28，and （时，分：9：30-16：00）
# trading_time_con = ((df0[i, 6] >= strat_time[0]) and (df0[i, 6] <= end_time[0])) and \
# ((df0[i, 7] == 9) and (df0[i, 8] >= 15)) or \
# ((df0[i, 7] == 16) and (df0[i, 8] <= 29)) or \
# ((df0[i, 7] > 9) and (df0[i, 7] < 16))
# # 我们交易时间
# #日
# we_trade_con = ((df0[i, 6] >= strat_time[0]) and (df0[i, 6] <= end_time[0])) and \
# ((df0[i, 7] == strat_time[1]) and (df0[i, 8] > strat_time[2])) or \
# ((df0[i, 7] == end_time[1]) and (df0[i, 8] < end_time[2])) or \
# ((df0[i, 7] > strat_time[1]) and (df0[i, 7] < end_time[1]))
# 交易所日盘，开放
if trading_time_con:
# 快捷变量
open_bar = df0[i][1]
high_bar = df0[i][2]
low_bar = df0[i][3]
last_close = df0[i - 1][4]
clsoe_bar = df0[i][4]
atr_n = df2[i-1][8] * atrn
swin_atr_n = df2[i-1][8] * sw_atrn
ma_xiao = df2[i][7]
da_ma = df2[i][10]
# === 仓位统计。
df1, now_open_prcie, now_pos = cal_per_pos(i, df1, open_pos_size, open_bar, clsoe_bar, last_close, sxf=1, slip=1)
# === 当日变量加载
if ((df0[i, 7] == 9) and (df0[i, 8] == 15)):
inner_trade_nums = trade_inner_nums
max_log_high = high_bar
min_log_low = low_bar
else:
max_log_high = max(max(open_bar,clsoe_bar), max_log_high)
min_log_low = min(min(open_bar,clsoe_bar), min_log_low)
inner_low = df2[i - 1][3]
stop_loss_price = (now_open_prcie + atr_n)
stop_win_price = now_open_prcie - stop_win_n
moving_swin_price = np.nan
# 信号生成
if we_trade_con:
if now_pos == 0:
short_condition = (clsoe_bar < inner_low) and (last_close >= inner_low) and (clsoe_bar<da_ma)
# 空：突破最低线。
if short_condition and inner_trade_nums > 0:
inner_trade_nums -= 1
df1[i][1] = -1
elif now_pos < 0:
stop_loss_con = (clsoe_bar > stop_loss_price) and\
(last_close <= stop_loss_price)
if min_log_low <stop_win_price:
stop_win_con = (clsoe_bar > ma_xiao)
moving_swin_price = ma_xiao
else:
stop_win_con = (clsoe_bar > ma_xiao + swin_atr_n)
moving_swin_price = ma_xiao+swin_atr_n
close_pos = False
if close_pos:
df1[i][1] = 0
elif stop_loss_con:
df1[i][1] = 0
elif stop_win_con:
df1[i][1] = 0
else: # 非交易时间段
if now_pos != 0:
df1[i][1] = 0
# 记录指标 绘图等等
# ['开仓线','atr_day','小均线']
df2[i, 1] = stop_loss_price
df2[i, 2] = stop_win_price
df2[i, 3] = min_log_low
df2[i, 4] = max_log_high
df2[i, 9] = moving_swin_price
res0 = cal_tongji(df_input=df1)
res0 = np.concatenate((res0, cs0))
return df0, df1, df2, res0
def cal_signal(df, strat_time, end_time, canshu):
a = time.process_time()
# ===转化成np.array
df0cols = ['candle_begin_time', 'open', 'high', 'low', 'close', 'volume', 'days', 'huors', 'minutes']
df1cols = ['candle_begin_time', 'signal', 'pos', 'opne_price', 'per_lr', 'sl']
df2cols = ['candle_begin_time', '止损价', '止盈价', '日内最低价', '日内最高价', '开仓线',
'atr_day', '小均线', 'n_atr','移动止赢','大均线']
# 配置参数:止盈，ma_len（atr），atr_倍数，交易次数
# stop_win_n, ma_len, atrn, trade_inner_nums = cs0
df['小均线'] = talib.SMA(df['close'], int(canshu[1]))
df['大均线'] = talib.SMA(df['close'], int(canshu[4]))
print(df)
df['n_atr'] = talib.ATR(df['close'], df['high'], df['low'], int(canshu[1]))
df['移动止赢'] = np.nan
# print(df.tail())
# exit()
df0 = df[df0cols].values
df1 = df[df1cols].values
df2 = df[df2cols].values
df0, df1, df2, res = cal_signal_(df0, df1, df2, strat_time, end_time, canshu)
print('runingtime:', time.process_time() - a, 's')
return df0, df1, df2, (df0cols, df1cols, df2cols), res
def duojincheng_backtesting(df_input, zong_can, strat_time, end_time, cpu_nums=3,jiexi=False):
df = df_input.copy()
if cpu_nums > cpu_count() - 1: cpu_nums = cpu_count() - 1
huice_df = []
def tianjia(res):
huice_df.append(res[-1])
# print(res[-1])
pass
# ===原始数据处理
df['candle_begin_time'] = (df['candle_begin_time'] - np.datetime64(0, 's')) / timedelta(seconds=1)
df['days'] = pd.to_datetime(df["candle_begin_time"], unit='s').apply(lambda x: float(x.to_pydatetime().day))
df['huors'] = pd.to_datetime(df["candle_begin_time"], unit='s').apply(lambda x: float(x.to_pydatetime().hour))
df['minutes'] = pd.to_datetime(df["candle_begin_time"], unit='s').apply(lambda x: float(x.to_pydatetime().minute))
# ===信号仓位统计数据列
df['signal'] = np.nan
df['pos'] = 0
df['opne_price'] = np.nan
df['per_lr'] = np.nan
df['sl'] = np.nan
# ===指标数据
df['止损价'] = np.nan
df['止盈价'] = np.nan
df['日内最低价'] = np.nan
df['日内最高价'] = np.nan
df['开仓线'] = np.nan
df['小均线'] = np.nan
df['大均线'] = np.nan
df = df[(df['huors'] >= 2) & (df['huors'] <= 16) & (df['days'] <= end_time[0])]
pass
# 多进程回测
if cpu_nums > 1:
p = Pool(processes=cpu_nums)
for j in range(0, len(zong_can), cpu_nums):
for i in range(cpu_nums):
if j + i <= len(zong_can) - 1:
canshu0 = zong_can[j + i]
p.apply_async(cal_signal, args=(df, strat_time, end_time, np.array(canshu0),), callback=tianjia, )
else:
break
p.close()
p.join()
print('进程池joined')
# 整理多进程回测的数据
cols = ['最后收益', '最大收益', '模拟最大回撤', '赢撤率', '夏普率', '平均收益', '开仓次数', '胜率', '盈亏比'] + [f'参数{i}' for i in range(1, len(zong_can[0]) + 1)]
resdf = pd.DataFrame(huice_df, columns=cols)
resdf = resdf[cols]
resdf.sort_values(by='最后收益', inplace=True)
print(resdf.iloc[-20:])
print(f'=参数回测结束,谢谢使用.')
return resdf
# 单进程测试
else:
for cs0 in zong_can:
a = time.process_time()
df0, df1, df2, cols, res0 = cal_signal(df, strat_time, end_time, np.array(cs0))
print('runingtime:', time.process_time() - a, 's')
tianjia(res=res0)
if jiexi :
return
df0cols, df1cols, df2cols = cols
# 转换成df数据
df00 = pd.DataFrame(df0, columns=df0cols)
df11 = pd.DataFrame(df1, columns=df1cols)
df22 = pd.DataFrame(df2, columns=df2cols)
# 合并
df11_ = pd.merge(df00, df11, on="candle_begin_time", suffixes=('_0', '_1'))
dfres = pd.merge(df11_, df22, on="candle_begin_time", suffixes=('', '_2'))
dfres["candle_begin_time"] = pd.to_datetime(dfres["candle_begin_time"], unit='s')
dfres.sort_values(by='candle_begin_time', inplace=True)
cols = ['最后收益', '最大收益', '模拟最大回撤', '赢撤率', '夏普率', '平均收益', '开仓次数', '胜率', '盈亏比'] + [f'参数{i}' for i in range(1, len(zong_can[0]) + 1)]
res0 = pd.DataFrame([res0], columns=cols)
print(dfres.iloc[-20:])
return dfres, res0
if __name__ == '__main__':
import time, datetime
import os
filename = os.path.abspath(__file__).split('\\')[-1].split('.')[0]
timeer0 = datetime.datetime.fromtimestamp(time.time()).strftime('%m_%d=%H_%M')
datapath = r'F:\task\恒生股指期货\hsi_data_1min\HSI2011-2019_12.csv'
dir_path =r'F:\task\恒生股指期货\numba_策略开发\numba_突破max01_突破_空策略'
# 1.先测试单进程 cpu == 1
a = 1
b = 0
c = 0
if True == a:
df_time_list = [['2019-2-1 09:15:00', '2019-12-10 16:25:00']]
s_time, e_time = df_time_list[0]
df = get_loc_hsicsv(s_time, e_time, datapath) # 获取本地数据
df000 = df.copy()
strat_time = np.array([1, 9, 30])
end_time = np.array([27, 16, 10])
time0 = time.process_time()
df000 = df.copy()
df000['ma']=talib.SMA(df000['close'],5000)
# df000['ma2']=talib.MA(df000['close'],5000)
print(df000)
# exit()
# resdf0,res01 = duojincheng_backtesting(df000.iloc[:1000], zong_can=[[80,180,200,0]], strat_time=strat_time, end_time=end_time, cpu_nums = 1)
# resdf0, res00 = duojincheng_backtesting(df000.iloc[:3000],
# zong_can=[[230, 16, 3,1]],
# strat_time=strat_time,
# end_time=end_time, cpu_nums=1,jiexi=True)
# time.sleep(0.2)
resdf, res0 = duojincheng_backtesting(df000, zong_can=[[210.0, 10.0, 4.0, 4.0, 5000.0, 1.0]],
strat_time=strat_time,
end_time=end_time, cpu_nums=1,jiexi=False)
print(resdf[resdf['signal']==-1])
print(resdf[resdf['signal']==-1].shape)
print(res0)
resdf['atr_day'] = resdf['日内最低价'] + resdf['atr_day']
mode = 112
if mode == 1:
resdf_baitian = resdf[(resdf['huors'] >= 2) & (resdf['huors'] <= 16)]
resdf = resdf_baitian.copy()
resdf.reset_index(inplace=True)
resdf.sort_values(by='candle_begin_time', inplace=True)
draw_charts(resdf, canshu={'opne_price': resdf['opne_price'],
'止损价': resdf['止损价'],
'止盈价': resdf['止盈价'],
'日内最低价': resdf['日内最低价'],
'日内最高价': resdf['日内最高价'],
'开仓线': resdf['开仓线'],
'小均线': resdf['小均线'],
'大均线': resdf['大均线'],
'移动止赢' :resdf['移动止赢'],
"atr_day": resdf['atr_day']
}, vol_bar=True, markline_show2=True,
path=r'html_gather\%s_%s' %(filename, timeer0))
if mode == 2:
resdf = resdf[resdf['pos'].notnull()].copy()
draw_charts(resdf, canshu={'opne_price': resdf['opne_price'],
'止损价': resdf['止损价'],
'止盈价': resdf['止盈价'],
'日内最低价': resdf['日内最低价'],
'日内最高价': resdf['日内最高价'],
'小均线': resdf['小均线'],
'大均线': resdf['大均线'],
'移动止赢' :resdf['移动止赢'],
"atr_day": resdf['atr_day']
}, vol_bar=True, markline_show2=True,
path=r'html_gather\%s_%s' %(filename, timeer0))
print(f'总时间：{time.process_time() - time0} s')
exit()
# 多进程策略 :
# [190, 20, 2, 2]
if True == b:
canshu_list = []
for cs1 in range(200, 280, 10):
for cs2 in range(10, 20, 3):
for cs3 in range(2, 5, 1):
for cs4 in range(1, 3, 1):
canshu_list.append([cs1, cs2, cs3, cs4])
canshu_list = canshu_list[:]
df_time_list = [['2016-1-10 09:15:00', '2019-12-20 16:25:00']]
s_time, e_time = df_time_list[0]
df = get_loc_hsicsv(s_time, e_time, datapath) # 获取本地数据
df000 = df.copy()
strat_time = np.array([1, 9, 30])
end_time = np.array([27, 16, 10])
print('参数列表个数：', len(canshu_list))
time0 = time.process_time()
resdf = duojincheng_backtesting(df000, zong_can=canshu_list[:], strat_time=strat_time, end_time=end_time, cpu_nums=2)
print(resdf.tail())
resdf.to_csv(r'csv_gather\%s_%s.csv' % (filename,timeer0))
print(f'总时间：{time.process_time() - time0} s')
# 多进程策略 :
if True == c:
from numba_策略开发.回测工具.多阶段回测 import duojieduan_huice
canshu_list = []
for cs1 in range(180, 280, 10):
for cs2 in range(10, 25, 5):
for cs3 in range(2, 5, 1):
for cs4 in range(2, 5, 1):
for cs5 in range(3000, 5001, 1000):
for cs6 in range(1, 3, 1):
canshu_list.append([cs1, cs2, cs3, cs4,cs5,cs6])
canshu_list = canshu_list[:]
df_time_list = [['2016-1-10 09:15:00', '2019-12-28 16:25:00']]
s_time, e_time = df_time_list[0]
df = get_loc_hsicsv(s_time, e_time, datapath) # 获取本地数据
df000 = df.copy()
strat_time = np.array([1, 9, 30])
end_time = np.array([27, 16, 10])
name = f'{filename}_{s_time[:4]}_{e_time[:4]}'
if True == 1:
print('参数列表个数：', len(canshu_list))
time0 = time.process_time()
func_canshu = {'zong_can': canshu_list[:], 'cpu_nums': 2, "strat_time": strat_time, "end_time": end_time}
df_zong = duojieduan_huice(df, duojincheng_backtesting, func_canshu, s_time, e_time, jiange='12MS')
path_csv = dir_path + r'\csv_gather\%s.csv'%name
df_zong.to_csv(path_csv)
# 接下来生成html
df_zong_html = pd.DataFrame()
for k, v in df_zong.groupby('s_time'):
print(k)
dfv = pd.DataFrame(v)
dfv.sort_values('最后收益', ascending=False, inplace=True)
df0 = dfv.iloc[:50]
df_zong_html = df_zong_html.append(df0, ignore_index=True)
canshu_cols = [i for i in df_zong_html.keys() if '参数' in i]
df_zong_html['参数_合并'] = df_zong_html[canshu_cols].values.tolist()
df_zong_html['参数_合并'] = df_zong_html['参数_合并'].astype(str)
df0 = df_zong_html['参数_合并'].value_counts() # .to_frame()
# df0.reset_index(inplace=True)
df_zong_html['参数出现次数'] = df_zong_html['参数_合并'].apply(lambda x: df0[x])
df_zong_html.set_index(keys=['s_time', '最后收益'], inplace=True)
html_path = dir_path+r'\策略介绍html\csv_html\%s.html' % name
df_zong_html.to_html(html_path)
# 读取本地数据,单独生成html，
if True == 0:
df_zong = pd.read_csv(dir_path + r'\csv_gather\%s.csv'%name)
df_zong_html = pd.DataFrame()
for k, v in df_zong.groupby('s_time'):
print(k)
dfv = pd.DataFrame(v)
dfv.sort_values('最后收益', ascending=False, inplace=True)
df0 = dfv.iloc[:50]
df_zong_html = df_zong_html.append(df0, ignore_index=True)
canshu_cols = [i for i in df_zong_html.keys() if '参数' in i]
df_zong_html['参数_合并'] = df_zong_html[canshu_cols].values.tolist()
df_zong_html['参数_合并'] = df_zong_html['参数_合并'].astype(str)
df0 = df_zong_html['参数_合并'].value_counts() # .to_frame()
# df0.reset_index(inplace=True)
df_zong_html['参数出现次数'] = df_zong_html['参数_合并'].apply(lambda x: df0[x])
df_zong_html.set_index(keys=['s_time', '最后收益'], inplace=True)
name = 'max_突破06_atr_2015_2019.html'
html_path = dir_path + r'\策略介绍html\csv_html\%s.html' % name
df_zong_html.to_html(html_path)
strat_time = np.array([1, 9, 30])
end_time = np.array([27, 16, 10])
name = f'{filename}_{s_time[:4]}_{e_time[:4]}'
if True == 1:
print('参数列表个数：', len(canshu_list))
time0 = time.process_time()
func_canshu = {
'zong_can': canshu_list[:],
'cpu_nums': 2,
"strat_time": strat_time,
"end_time": end_time
}
df_zong = duojieduan_huice(df,
duojincheng_backtesting,
func_canshu,
s_time,
e_time,
jiange='12MS')
path_csv = dir_path + r'\csv_gather\%s.csv' % name
df_zong.to_csv(path_csv)
# 接下来生成html
df_zong_html = pd.DataFrame()
for k, v in df_zong.groupby('s_time'):
print(k)
dfv = pd.DataFrame(v)
dfv.sort_values('最后收益', ascending=False, inplace=True)
df0 = dfv.iloc[:50]
df_zong_html = df_zong_html.append(df0, ignore_index=True)
canshu_cols = [i for i in df_zong_html.keys() if '参数' in i]
df_zong_html['参数_合并'] = df_zong_html[canshu_cols].values.tolist()
df_zong_html['参数_合并'] = df_zong_html['参数_合并'].astype(str)
if True == 1:
print('参数列表个数：', len(canshu_list))
time0 = time.process_time()
# df, strat_time, end_time, cs0, ix_zong = 100, all_day = 0
func_canshu = {
'zong_can': canshu_list[:],
'cpu_nums': 2,
"path": r"",
"all_day": 0,
"strat_time": strat_time,
"end_time": end_time
}
df_zong = duojieduan_huice(df,
cal_signal,
func_canshu,
s_time,
e_time,
jiange='12MS')
path_csv = dir_path + r'\csv_gather\%s.csv' % name
df_zong.to_csv(path_csv)
# 接下来生成html
df_zong_html = pd.DataFrame()
for k, v in df_zong.groupby('s_time'):
print(k)
dfv = pd.DataFrame(v)
dfv.sort_values('最后收益', ascending=False, inplace=True)
df0 = dfv.iloc[:50]
df_zong_html = df_zong_html.append(df0, ignore_index=True)
canshu_cols = [i for i in df_zong_html.keys() if '参数' in i]
df_zong_html['参数_合并'] = df_zong_html[canshu_cols].values.tolist()
df_zong_html['参数_合并'] = df_zong_html['参数_合并'].astype(str)
