def mserror(ts_id, win, **kwargs):
ts = dat.get_series(ts_id)[:, 0]
tsdf = pd.Series(ts)
bn = analysis.get_best_net(ts_id)
mse = lambda win: np.mean(win - bn.predict(
np.array(win, dtype='float32')[:, None, None]))**2
if win == 0: #no window. just return all errors at once
pr = (bn.predict(ts[:, None, None])[:, 0, 0] - ts)**2
return pr
return \
rolling_apply(tsdf,
win
,mse
,center=True
)
def env(ts_id, **kwargs):
"""use dbts_id='test' kwargs to test things"""
global gts_id
gts_id = kwargs.setdefault('dbts_id', ts_id)
global trn
global vld
global dim_out
global dim_in
global noise
ts = dat.get(ts_id)
tl = int(.75 * len(ts)) #potential <-param here
trn = dat.list_call(ts[:tl])
vld = dat.list_call(ts[tl:])
dim_out = dim_in = dat.dim(ts_id)
noise = np.std(dat.get_series(ts_id)) * .75 #<- critical param
def data(self):
er=[analysis.errs(self.name,awin) for awin in self.wins ]
tsd=dat.get_series(self.name)
return tsd,er