def fit(self, X, y):
"""Fit."""
X = X[(self.mdlNr*5 % self.subsample_data)::self.subsample_data]
y = y[(self.mdlNr*5 % self.subsample_data)::self.subsample_data]
if self.applyPreds:
if self.jump is not None:
X = delay_preds(X, delay=self.delay/self.subsample_data, skip=self.skip/self.subsample_data, jump=self.jump/self.subsample_data)
else:
X = delay_preds(X, delay=self.delay/self.subsample_data, skip=self.skip/self.subsample_data)
self.clf = []
widgets = ['Training : ', Percentage(), ' ', Bar(marker=RotatingMarker()),
' ', ETA(), ' ']
pbar = ProgressBar(widgets=widgets, maxval=6)
pbar.start()
# training separate models for each event
for col in range(6):
self.clf.append(xgb.XGBClassifier(n_estimators=self.n_estimators,
max_depth=self.max_depth,
subsample=self.subsample,
nthread=self.nthread))
self.clf[col].fit(X, y[:, col])
pbar.update(col)
def predict_proba(self, X):
"""Predict probability."""
if self.applyPreds:
p = np.zeros((X.shape[0],6))
for part in range(self.partsTest):
start = part*X.shape[0]//self.partsTest-self.delay*(part>0)
stop = (part+1)*X.shape[0]//self.partsTest
X_delayed = delay_preds(X[slice(start,stop)], delay=self.delay, skip=self.skip, jump=self.jump)[self.delay*(part>0):]
start += self.delay*(part>0)
p[slice(start,stop)] += self._predict_proba(X_delayed)
X_delayed = None
return p
else:
return self._predict_proba(X)