IRLS的简单例子

from numpy import array, diag, dot, maximum, empty, repeat, ones, sum
from numpy.linalg import inv
from sklearn import datasets
from sklearn.linear_model import LinearRegression
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error

def IRLS(X, y, maxiter, w_init=1, d=0.0001, tol=0.0001):
    nSample, nDim = X.shape
    delta = array(repeat(d,nSample)).reshape(1,nSample)
    w = repeat(1,nSample)
    W = diag(w)
    B = inv(X.T @ W @ X) @ X.T @ W @ y
    for _ in range(maxiter):
        _B = B
        _w = abs(y-X @ B).T
        w = float(1.0) / maximum(delta, _w)
        W = diag(w[0])
        B = inv(X.T @ W @ X) @ X.T @ W @ y
        if sum(abs(B-_B)) < tol:
            return B
    return B

# X, y = datasets.load_iris(return_X_y=True)
X, y = datasets.load_boston(return_X_y=True)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=3)

B = IRLS(X_train,y_train,maxiter=500)
y_hat = X_test @ B
MSE = mean_squared_error(y_true=y_test, y_pred=y_hat)
print("IRLS::",MSE)

model = LinearRegression()
model.fit(X_train, y_train)
y_hat = model.predict(X=X_test)
MSE = mean_squared_error(y_true=y_test, y_pred=y_hat)
print("LR::",MSE)



# print(abs(y - X @ B))

Python Implementation of Iterative Reweighted Least Square of Logistic Regression

https://github.com/jaeho3690/LogisiticRegression