def fit(self, X, y, *args, parallel=None, **kwargs):
"""
:param X: n_samples X n_models
:param y: n_samples
:param args: placeholder
:param kwargs: placeholder
"""
data = NpDataset(X, y=y)
self.__val_preds = np.zeros(y.shape)
if self.__val_preds.ndim != 1:
logging.error(
"Shape of validation predictions is incorrect: {}".format(
self.__val_preds.shape))
model_and_preds = parallel(
delayed(fit_fold_model)(self.models[i], train_data.x, train_data.y,
val_data.x, val_data.y, *args, **kwargs)
for i, (train_data,
val_data) in enumerate(data.kfold(self.k, shuffle=False)))
cur_sample_ind = 0
for i, (model, val_preds) in enumerate(model_and_preds):
self.models[i] = model
self.__val_preds[cur_sample_ind:cur_sample_ind +
val_preds.shape[0]] = val_preds
cur_sample_ind += val_preds.shape[0]
assert cur_sample_ind == X.shape[0]
def roc_auc(self, X, y, *args, **kwargs):
data = NpDataset(X, y=y)
score = 0.
for i, (train_data,
val_data) in enumerate(data.kfold(self.k, shuffle=False)):
score = score + self.models[i].roc_auc(val_data.x, val_data.y, *
args, **kwargs)
return score / self.k