3.3 softmax回归推导及python实现

from sklearn import datasets
from sklearn.model_selection import train_test_split
import numpy as np

data = datasets.load_iris()
X = data['data']
y = data['target']
# n:特征数，k:标签数
n = X.shape[1]
label = np.unique(y)
k = len(label)

X = np.insert(X, 0, 1, axis=1)  # 添加一列 m*n+1
X_train, X_test, Y_train, Y_test = train_test_split(X, y, train_size=0.75, test_size=0.15, random_state=1)

# m:样本数
m = X_train.shape[0]


def getGradient(theta):
    # 生成梯度矩阵G
    G = np.zeros((k, m))  # k*m
    for i in range(m):
        G[Y_train[i]][i] = 1
    # 生成假设矩阵H
    H = hypothesis(theta, X_train)
    # 计算梯度
    return np.dot((G - H), X_train)


def hypothesis(theta, X):
    eta = np.dot(theta, X.T)  # k*m
    H = np.exp(eta)  # k*m
    H /= np.sum(H, axis=0)
    return H


d