一、引言
对于单模型来说，模型的抗干扰能力低，且难以拟合复杂的数据。
所以可以集成多个模型的优缺点，提高泛化能力。
集成学习一般有三种：boosting是利用多个弱学习器串行，逐个纠错，构造强学习器。
bagging是构造多个独立的模型，然后增强泛化能力。
而stacking结合了以上两种方式，将xy先进行n-fold，然后分给n个基学习器学习，再将n个输出的预测值进行堆叠，形成新的样本数据作为x。新的x和旧的y交给第二层模型进行拟合。
二、代码
import numpy as np
from sklearn.model_selection import KFold
from sklearn import datasets
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
class MyStacking:
初始化模型参数
def init(self, estimators, final_estimator, cv=5, method='predict'):
self.cv = cv
self.method = method
self.estimators = estimators
self.final_estimator = final_estimator
模型训练
def fit(self, X, y):
获得一级输出
dataset_train = self.stacking(X, y)
模型融合
self.final_estimator.fit(dataset_train, y)
堆叠输出
def stacking(self, X, y):
kf = KFold(n_splits=self.cv, shuffle=True, random_state=2021)
获得一级输出
dataset_train = np.zeros((X.shape[0], len(self.estimators)))
for i, model in enumerate(self.estimators):
for (train, val) in kf.split(X, y):
X_train = X[train]
X_val = X[val]
y_train = y[train]
y_val_pred = model.fit(X_train, y_train).predict(X_val)
dataset_train[val, i] = y_val_pred
self.estimators[i] = model
return dataset_train
模型预测
def predict(self, X):
datasets_test = np.zeros((X.shape[0], len(self.estimators)))
for i, model in enumerate(self.estimators):
datasets_test[:, i] = model.predict(X)
return self.final_estimator.predict(datasets_test)
模型精度
def score(self, X, y):
datasets_test = np.zeros((X.shape[0], len(self.estimators)))
for i, model in enumerate(self.estimators):
datasets_test[:, i] = model.predict(X)
return self.final_estimator.score(datasets_test, y)
if name == 'main':
X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(
X, y, train_size=0.7, random_state=0)
estimators = [
RandomForestClassifier(n_estimators=10),
GradientBoostingClassifier(n_estimators=10)
]
clf = MyStacking(estimators=estimators,
final_estimator=LogisticRegression())
clf.fit(X_train, y_train)
print(clf.score(X_train, y_train))
print(clf.score(X_test, y_test))