from typing import List
import numpy as np
from sklearn.base import BaseEstimator
from autoflow.data_container.base import get_container_data
from autoflow.utils import typing_
[docs]class EnsembleEstimator(BaseEstimator):
mainTask = None
[docs] def build_prediction_list(self):
prediction_list = []
assert len(self.y_true_indexes_list) > 1
# splitter 的 random_state都是相同的， 所以认为 y_true_indexes_list 的每个 y_true_indexes 都相同
assert not np.any(np.array([np.hstack(y_true_indexes) for y_true_indexes in self.y_true_indexes_list]).var(axis=0))
for y_preds in self.y_preds_list:
prediction_list.append(np.concatenate(y_preds)) # concat in axis 0
self.prediction_list = prediction_list
y_true_indexes = self.y_true_indexes_list[0]
self.stacked_y_true = self.y_true[np.hstack(y_true_indexes)]
assert self.prediction_list[0].shape[0] == self.stacked_y_true.shape[0]
[docs] def fit_trained_data(
self,
estimators_list: List[List[typing_.GenericEstimator]],
y_true_indexes_list: List[List[np.ndarray]],
y_preds_list: List[List[np.ndarray]],
y_true: np.ndarray
):
self.y_preds_list = y_preds_list
self.y_true_indexes_list = y_true_indexes_list
self.estimators_list = estimators_list
self.y_true = get_container_data(y_true)
self.build_prediction_list()