def ai_f_mxWrlst(ftg0,funlst,x_train, y_train):
for funSgn in funlst:
ftg=ftg0+funSgn+'.pkl'
print('\n',ftg)
tim0=arrow.now()
ai_f_mxWr(ftg,funSgn,x_train, y_train)
zt.timNSec('',tim0,True)
def main_get(timStr='', nday=2):
#
#1---init.sys
print('\nmain_get,nday:', nday)
tfsys.xnday_down = nday
zsys.web_get001txtFg = True
#
#2---init.tfb
rs0 = '/tfbDat/'
fgid = rs0 + 'gid2017.dat'
xtfb = tft.fb_init(rs0, fgid)
if nday == -1:
tfsys.xnday_down = xtfb.gid_nday + 10
print('nday,', tfsys.xnday_down)
#
#3---update.data
print('\n#3,update.data')
if nday != 0:
tft.fb_gid_get_nday(xtfb, timStr, fgExt=True)
#
#4
tn = zt.timNSec(timStr, xtfb.tim0, '')
print('\n#4,update.data,tim:{0:.2f} s'.format(tn))
def ai_mx_tst_bsize(f_mx,
f_tg,
df_train,
df_test,
nepochs=500,
ksize=32,
ky0=5):
ds, df = {}, pd.DataFrame()
for xc in range(1, 11):
print('\n#', xc)
dnum = xc * ksize
mx = ks.models.load_model(f_mx)
t0 = arrow.now()
dacc = ai_mul_var_tst(mx, df_train, df_test, nepochs, dnum, ky0=ky0)
tn = zt.timNSec('', t0)
ds['bsize'], ds['size_acc'], ds['ntim'] = dnum, dacc, tn
df = df.append(ds, ignore_index=True)
#
df = df.dropna()
df['bsize'] = df['bsize'].astype(int)
print('\ndf')
print(df)
print('\nf,', f_tg)
df.to_csv(f_tg, index=False)
#
df.plot(kind='bar', x='bsize', y='size_acc', rot=0)
df.plot(kind='bar', x='bsize', y='ntim', rot=0)
return df
def mx_funlst8mx(mxlst, x_test, y_test,yk0=5,fgInt=False):
for msgn in mxlst:
#print('@msgn:',msgn)
tim0=arrow.now()
mx=xmodel[msgn]
dacc,df9=mx_fun8mx(mx, x_test, y_test,yk0,fgInt)
tn=zt.timNSec('',tim0)
xss='ok:{0:.2f}%,mx,{1},tn,{2:.2f} s'.format(dacc,msgn,tn)
print(xss)
def main_ai_bt(timStr='', nday=2):
#
#1---init.sys
print('\nmain_bt,nday:', nday)
tfsys.xnday_down = nday
zsys.web_get001txtFg = True
#2---init.tfb
rs0 = '/tfbDat/'
fgid = rs0 + 'gid2018.dat'
xtfb = tft.fb_init(rs0, fgid)
if nday == -1:
tfsys.xnday_down = xtfb.gid_nday + 10
print('nday,', tfsys.xnday_down)
#
#3---backtest
print('\n#3,backtest')
if nday != 0:
xtfb.funPre = tfsty.sta00_pre #bt_1dayMain
xtfb.funSta = tfsty.sta_ai_log01
xtfb.preVars = []
xtfb.staVars = [99, 99, 99]
#
#
#3.a------ai.init
zai.xmodel = {}
xtfb.ai_mxfFN0 = rs0 + 'mlib/p7y2016_'
xtfb.ai_mx_sgn_lst = ['log']
xtfb.ai_ysgn = 'kwin'
xtfb.ai_xlst = [
'cid', 'pwin0', 'pdraw0', 'plost0', 'pwin9', 'pdraw9', 'plost9'
]
#
zai.ai_f_mxRdlst(xtfb.ai_mxfFN0, xtfb.ai_mx_sgn_lst)
#3.b-------预测
xtfb.kcid = '1' #cn,3=bet365
tfbt.bt_main(xtfb, timStr)
#
#4---main_ret
print('\n#4,result.anz')
tfbt.bt_main_ret(xtfb, True)
print('kcid,', xtfb.kcid, ',nday,', nday)
print('preVar,', xtfb.preVars)
print('staVar,', xtfb.staVars)
#
#5
tn = zt.timNSec('', xtfb.tim0, '')
print('\n#5,backtest,tim:{0:.2f} s'.format(tn))
#
#6---end.main
print('\n#6,end.main')
def main_bt(timStr='',nday=2):
#
#1---init.sys
print('\nmain_bt,nday:',nday)
tfsys.xnday_down=nday
# zsys.web_get001txtFg= True
#2---init.tfb
rs0='/tfbDat/'
fgid=rs0+'gid2019_js.dat'
xtfb=tft.fb_init(rs0,fgid)
if nday==-1:
tfsys.xnday_down=xtfb.gid_nday+2
print('nday,',tfsys.xnday_down)
#
#3---backtest
print('\n#3,backtest')
if nday!=0:
xtfb.funPre=tfsty.sta_lstm_pre ##预测算法
xtfb.funSta=tfsty.sta00_sta
xtfb.preVars=[]
xtfb.staVars=[1.5]
xtfb.kcid='1' #cn,3=bet365
#
tfbt.bt_main(xtfb,timStr)
print('\n#4, 完成预测')
#4---main_ret 按预测计算预计回报
#暂时不分析预计回报，以后再开发
'''
print('\n#4,result.anz')
tfbt.bt_main_ret(xtfb, True)
print('kcid,',xtfb.kcid,',nday,',nday)
print('preVar,',xtfb.preVars)
print('staVar,',xtfb.staVars)
'''
#
#5
tn=zt.timNSec('',xtfb.tim0,'')
print('\n#5,backtest,tim:{0:.2f} s'.format(tn))
#
#6---end.main
print('\n#6,end.main')
def mx_mul(mlst, x_test, y_test, yk0=5, fgInt=False, fgDebug=False):
#1
print('\ny_pred,预测')
df9, xc, mxn9 = x_test.copy(), 0, len(mlst)
df9['y_test'] = y_test
#2
for msgn in mlst:
xc += 1
tim0 = arrow.now()
mx = xmodel[msgn]
y_pred = mx.predict(x_test.values)
#3
if xc == 1: df9['y_sum'] = y_pred
else: df9['y_sum'] = df9['y_sum'] + y_pred
#4
tn = zt.timNSec('', tim0)
df9['y_pred'] = y_pred
#4.b
if fgInt:
df9['y_predsr'] = df9['y_pred']
df9['y_pred'] = round(df9['y_predsr']).astype(int)
#5
dacc = ai_acc_xed(df9, yk0, fgDebug)
xss = 'y_pred{0:02},kok:{1:.2f}%'.format(xc, dacc)
print(xc, xss, msgn, tn, 's')
ysgn = 'y_pred' + str(xc)
df9[ysgn] = y_pred
#6
df9['y_pred'] = df9['y_sum'] / mxn9
if fgInt:
df9['y_predsr'] = df9['y_pred']
df9['y_pred'] = round(df9['y_predsr']).astype(int)
#7
dacc = ai_acc_xed(df9, yk0, fgDebug)
#8
if fgDebug:
df9.to_csv('tmp/df9_pred.csv')
#9
print('@mx:mx_sum,kok:{0:.2f}%'.format(dacc))
return dacc, df9
#
mx.summary()
plot_model(mx, to_file='tmp/lstm010mx.png')
#4 模型训练
print('\n#4 模型训练 fit')
tbCallBack = keras.callbacks.TensorBoard(log_dir=rlog,
write_graph=True,
write_images=True)
tn0 = arrow.now()
mx.fit(x_train,
y_train,
epochs=2000,
batch_size=512 * 24,
callbacks=[tbCallBack])
tn = zt.timNSec('', tn0, True)
mx.save('tmp/lstm010mx.dat')
#5 利用模型进行预测 predict
print('\n#5 模型预测 predict')
tn0 = arrow.now()
y_pred = mx.predict(x_test)
tn = zt.timNSec('', tn0, True)
df_test['y_pred'] = zdat.ds4x(y_pred, df_test.index, True)
df_test.to_csv('tmp/df_lstm010mx.csv', index=False)
#6
print('\n#6 acc准确度分析')
print('\nky0=10')
df = df_test
dacc, dfx, a10 = ztq.ai_acc_xed2ext(df.y, df.y_pred, ky0=10, fgDebug=True)
import zpd_talib as zta
import ztop_ai as zai
#-----------------------
#1
fsr0 = 'dat/ccpp_'
print('#1', fsr0)
xlst, ysgn = ['AT', 'V', 'AP', 'RH'], 'PE'
x_train, x_test, y_train, y_test = zai.ai_dat_rd(fsr0)
funSgn, ftg = 'svm', 'tmp/ccpp_svm.pkl'
#2
print('\n#2,mx_svm.wr')
tim0 = arrow.now()
zai.ai_f_mxWr(ftg, funSgn, x_train, y_train)
tn = zt.timNSec('', tim0, True)
#3
tim0 = arrow.now()
print('\n#3,mx_svm.rd')
mx = joblib.load(ftg)
tn = zt.timNSec('', tim0, True)
#4
print('\n#4,mx_svm')
tim0 = arrow.now()
zai.mx_fun8mx(mx, x_test, y_test)
tn = zt.timNSec('', tim0, True)
#-----------------------
print('\nok!')
def fun_tim(*args, **kwargs):
t0 = arrow.now()
result = function(*args, **kwargs)
tn = zt.timNSec(arrow.now(), t0)
print('tn,{0:.3f}s,fun:{1}'.format(tn, function.__name__))
return result
def mx_funlst(funlst,x_train, x_test, y_train, y_test,yk0=5,fgInt=False):
for funsgn in funlst:
print('\n',funsgn)
tim0=arrow.now()
mx_fun010(funsgn,x_train, x_test, y_train, y_test,yk0,fgInt)
zt.timNSec('',tim0,True)
