def total(self, dfs, kt):
#uid = self.uidKey % ( '_'.join(self.codes), self.klineType[:-1])
df0, df1 = dfs[self.mCodes[0]], dfs[self.mCodes[1]]
df0 = df0.dropna(axis=0)
df1 = df1.dropna(axis=0)
if len(df0) != len(df1):
#print('------------------change', kt, len(df0), len(df1))
if kt[-1] == 'm':
if len(df0) > len(df1):
df0 = df0[df0.index.isin(df1.index)]
df1 = df1[df1.index.isin(df0.index)]
else:
df1 = df1[df1.index.isin(df0.index)]
df0 = df0[df0.index.isin(df1.index)]
elif len(df0) > len(df1):
df0 = df0[df1.index[0]:]
else:
df1 = df1[df0.index[0]:]
print(len(df0), len(df1))
c0, c1 = df0['close'], df1['close']
# 交易量差距
volrate = df0['volume'].mean() / df1['volume'].mean()
# 涨跌一致性
diff0 = c0.diff().apply(lambda x: 1 if x > 0 else -1 if x<0 else 0)
diff1 = c1.diff().apply(lambda x: 1 if x > 0 else -1 if x<0 else 0)
df0 = df0.dropna(axis=0)
df1 = df1.dropna(axis=0)
ss = ((diff0+diff1)).apply(lambda x: 1 if x!=0 else 0)
delta = ss.sum() / len(ss)
# 波动比
s0 = ta.MA(c0,timeperiod=15).mean()
sd0 = ta.STDDEV(c0, timeperiod=15, nbdev=1).mean()
s1 = ta.MA(c1, timeperiod=15).mean()
sd1 = ta.STDDEV(c1, timeperiod=15, nbdev=1).mean()
std = sd0 * s1 / s0 / sd1
#
# 跌涨标准差
dd0 = ta.STDDEV(c0.diff(), timeperiod=15, nbdev=1).mean()
dd1 = ta.STDDEV(c1.diff(), timeperiod=15, nbdev=1).mean()
diffstd = dd0 * s1 / s0 / dd1
# 协整
coint = sm.tsa.stattools.coint(c0, c1)
# x相关性
relative = per(c0, c1)
doc = {
"code": self.codes[0],
"code1": self.codes[1],
"kline": self.klineType,
"relative": relative,
"samecount": len(df0),
"samerate": delta,
"diffstd": diffstd,
"coint_1": 0 if np.isnan(coint[0]) else coint[0],
"coint": 0 if np.isnan(coint[1]) else coint[1],
"std": std,
"vol": volrate,
"type": "future"
}
print(doc)
# 按时间截取并调整
print('kline load:', kt, self.codes, len(df0), len(df1))
return doc
def total(self, dfs, dfs2=None, scale=1, period=60):
uid = self.uidKey % ("_".join(self.codes), str(period), str(scale),
self.klineType[:-1], str(self.shiftScale))
df0, df1 = dfs[self.mCodes[0]], dfs[self.mCodes[1]]
df0["rel_price"] = close = df0["close"] / df1["close"]
df0["datetime"] = df0.index
# 模拟滑点
num = len(df0)
s0, s1 = self.shift[0], self.shift[1]
p_l = df0["p_l"] = (df0["close"] + s0) / (df1["close"] - s1)
p_h = df0["p_h"] = (df0["close"] - s0) / (df1["close"] + s1)
close2 = dfs2[self.mCodes[0]]["close"] / dfs2[self.mCodes[1]]["close"]
#ma = ta.MA(close, timeperiod=period)
#df0["rel_std"] = sd = ta.STDDEV(close, timeperiod=period, nbdev=1)
df0['ma'] = ma = df0.apply(lambda row: self.k_ma(
row['datetime'], row['rel_price'], close2, period, 0),
axis=1)
df0['rel_std'] = sd = df0.apply(lambda row: self.k_ma(
row['datetime'], row['rel_price'], close2, period, 1),
axis=1)
# 上下柜
top, lower = ma + scale * sd, ma - scale * sd
# bullWidth
df0["bullwidth"] = width = (4 * sd / ma * 100).fillna(0)
# 近三分钟width变动
df0["widthDelta"] = ta.MA(width - width.shift(1),
timeperiod=3).fillna(0)
df0["delta"] = (p_l - p_h) / sd
# 其他参数计算
min, max = ta.MINMAX(close, timeperiod=period)
df0["atr"] = ta.WMA((max.dropna() - min.dropna()),
timeperiod=period / 2)
# 协整
result = sm.tsa.stattools.coint(df0["close"], df1["close"])
df0.fillna(0, inplace=True)
isOpen, preDate, prePrice = 0, None, 0
doc, doc1, docs = {}, {}, []
for i in range(num):
isRun = False
if i < period and np.isnan(ma[i]): continue
# 开仓
if isOpen == 0:
# 大于上线轨迹
if p_h[i] >= top[i]:
isOpen = -1
isRun = True
elif p_l[i] <= lower[i]:
isOpen = 1
isRun = True
# 平仓
else:
# 回归ma则平仓 或 超过24分钟 或到收盘时间 强制平仓
if (isOpen *
((p_h[i] if isOpen == 1 else p_l[i]) - ma[i])) >= 0:
isOpen = 0
isRun = True
# 止损
if isRun:
doc, doc1 = self.order(df0.iloc[i], df1.iloc[i], isOpen, uid)
if doc1 is not None:
docs.append(doc)
docs.append(doc1)
res = pd.DataFrame(docs)
#res.fillna(0,inplace=True)
if len(res) > 0:
if self.saveDetail:
print(self.Train.tablename, len(res), 'save')
self.Train.insertAll(docs)
return {
"scale":
scale,
"code":
self.codes[0],
"code1":
self.codes[1],
"period":
period,
"count":
int(len(docs) / 4),
"amount":
(doc["price"] * doc["vol"] +
doc1["price"] * doc1["vol"]) if doc is not None else 0,
"price":
doc["price"] / doc1["price"],
"income":
res["income"].sum(),
"uid":
uid,
"relative":
per(df0["close"], df1["close"]),
"std":
res['rel_std'].mean(),
"shift": (p_l - p_h).mean(),
"delta":
res['delta'].mean(),
"coint":
0 if np.isnan(result[1]) else result[1],
"createdate":
public.getDatetime()
}
else:
return None
def total(self, dfs, dfs2=None, period=60):
# 计算参数
df0, df1 = dfs[self.mCodes[0]], dfs[self.mCodes[1]]
df0["rel_price"] = close = df0["close"] / df1["close"]
df0["datetime"] = df0.index
s0, s1 = self.shift[0], self.shift[1]
p_l = df0["p_l"] = (df0["close"] + s0) / (df1["close"] - s1)
p_h = df0["p_h"] = (df0["close"] - s0) / (df1["close"] + s1)
if self.isSimTickUse:
# 调用复合apply函数计算混合参数
close2 = dfs2[self.mCodes[0]]["close"] / dfs2[self.mCodes[1]]["close"]
df0_1 = df0.apply(lambda row: self.k_ma(row['datetime'], row['rel_price'], close2, period), axis=1)
df0 = pd.concat([df0, df0_1], axis=1)
else:
df0["ma"] = ma = ta.MA(close, timeperiod=period)
df0["std"] = std = ta.STDDEV(close, timeperiod=period, nbdev=1)
# 上下柜
# bullWidth
df0["bullwidth"] = width = (4 * std / ma * 100).fillna(0)
# 近三分钟width变动
df0["widthDelta"] = wd1 = ta.MA(width - width.shift(1), timeperiod=3).fillna(0)
df0["widthDelta2"] = wd2 = wd1 - wd1.shift(1)
df0["wd2m"] = wd2 * wd2.shift(1)
# 相对波动
df0['delta'] = (p_l - p_h) / df0['std']
# 协整
coint = sm.tsa.stattools.coint(df0["close"], df1["close"])
# x相关性
relative = per(df0["close"], df1["close"])
# 循环 scale
docs = []
for scale in self.scaleList:
for conds in self.iterCond():
uid = self.uidKey % (
'_'.join(self.codes), str(period), str(scale), self.klineType[:-1], conds)
self.stopTimeDiff = self.stopTimeLine * period * int(self.klineType[:-1])
# 计算高低线值
df0["top"], df0["lower"] = df0['ma'] + (scale-self.scaleDiff) * df0['std'], df0['ma'] - (scale+self.scaleDiff) * df0['std']
# df0.fillna(0, inplace=True)
tot = self.detect(df0, df1, period=period, uid=uid)
if tot is not None and tot['amount'] != 0:
tot.update(
{
"scale": scale,
"code": self.codes[0],
"code1": self.codes[1],
"period": period,
"uid": uid,
"relative": relative,
"shift": (p_l - p_h).mean(),
"coint": 0 if np.isnan(coint[1]) else coint[1],
"createdate": public.getDatetime()
}
)
docs.append(tot)
return docs
def total(self, df0, df1=None, scale=1, period=60):
uid = self.uid
df0["rel_price"] = close = df0["close"] / df1["close"]
df0["datetime"] = df0.index
s0, s1 = self.shift[0], self.shift[1]
p_l = df0["p_l"] = (df0["close"] + s0) / (df1["close"] - s1)
p_h = df0["p_h"] = (df0["close"] - s0) / (df1["close"] + s1)
num = len(df0)
ma = ta.MA(close[0:num], timeperiod=period)
df0["std"] = std = ta.STDDEV(close, timeperiod=period, nbdev=1)
# 上下柜
top, lower = ma + scale * std, ma - scale * std
# bullWidth
df0["bullwidth"] = width = (4 * std / ma * 100).fillna(0)
# 近三分钟width变动
df0["widthDelta"] = wd1 = ta.MA(width - width.shift(1),
timeperiod=3).fillna(0)
df0["delta"] = (p_l - p_h) / std
wd2 = wd1 - wd1.shift(1)
wd2m = wd2 * wd2.shift(1)
# 其他参数计算
min, max = ta.MINMAX(close, timeperiod=period)
df0["atr"] = ta.WMA((max.dropna() - min.dropna()),
timeperiod=period / 2)
# 协整
result = sm.tsa.stattools.coint(df0["close"], df1["close"])
df0.fillna(0, inplace=True)
isOpen, preDate, prePrice = 0, None, 0
doc, doc1, docs = {}, {}, []
sline, wline = self.stopTimeDiff, self.widthDeltaLine
for i in range(period, num):
isRun, isstop = False, 0
# 开仓2
if isOpen == 0:
cond1, cond2 = True, False
if wline > 0:
# 布林宽带变化率
cond1 = not ((wd1[i] > wline) or (wd2[i] > (wline / 2)))
# 最大值
cond2 = wd2m[i] < 0
# 突变状态开始
# 大于上线轨迹
if p_h[i] >= top[i] and not cond1:
isOpen = -1
isRun = True
elif p_l[i] <= lower[i] and not cond1:
isOpen = 1
isRun = True
elif p_h[i] >= top[i] and cond1 and cond2:
isOpen = -2
isRun = True
elif p_l[i] <= lower[i] and cond1 and cond2:
isOpen = 2
isRun = True
# 平仓
else:
# 回归ma则平仓 或 超过24分钟 或到收盘时间 强制平仓
if (isOpen *
((p_h[i] if isOpen == 1 else p_l[i]) - ma[i])) >= 0:
isOpen = 0
isRun = True
# 止损
elif sline > 0 and self.preNode and len(self.preNode) == 2:
timediff = public.timeDiff(
df0['datetime'].values,
self.preNode[0]['createdate']) / 60
if timediff > sline:
isOpen, isstop = 0, 1
isRun = True
if isRun:
doc, doc1 = self.order(df0.iloc[i],
df1.iloc[i],
isOpen,
uid,
isstop=isstop)
if doc1 is not None:
docs.append(doc)
docs.append(doc1)
res = pd.DataFrame(docs).fillna(0).replace(np.inf,
0).replace(-np.inf, 0)
if self.saveDetail:
self.Train.insertAll(res.to_dict(orient='records'))
if len(res) > 0:
return {
"scale":
scale,
"code":
self.codes[0],
"code1":
self.codes[1],
"period":
period,
"count":
int(len(docs) / 4),
"amount":
(doc["price"] * doc["vol"] +
doc1["price"] * doc1["vol"]) if doc is not None else 0,
"price":
doc["price"] / doc1["price"],
"income":
res["income"].sum(),
"uid":
uid,
"relative":
per(df0["close"], df1["close"]),
"std":
std.mean(),
"shift": (p_l - p_h).mean(),
"delta": (p_l - p_h).mean() / std.mean(),
"coint":
0 if np.isnan(result[1]) else result[1],
"createdate":
public.getDatetime()
}
else:
return None