模型评估-K折交叉验证法

点击查看代码

from sklearn import datasets  # 自带数据集
from sklearn.model_selection import train_test_split  # 数据集划分
from sklearn.preprocessing import StandardScaler  # 标准化
import numpy as np
import matplotlib.pyplot as plt
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import cross_val_score

iris = datasets.load_iris()
# iris
X, target = iris.data, iris.target

def knn_model_train(X, y):
    # 采用诗词交叉验证法对k近邻算法进行超参数选择
    X_train, X_test, y_train, y_test = train_test_split(X, target, test_size=0.3,
                                                        random_state=111, shuffle=True, stratify=y)
    k_range = range(1, 31)  # 超参数选择区间
    cv_scores = []  # 存储每次调参的10折交叉验证精度均值
    for k in k_range:
        knn = KNeighborsClassifier(k)  # k值
        # 聚类算法的指标 "accuracy"
        # https://scikit-learn.org/stable/modules/model_evaluation.html#scoring-parameter
        scores = cross_val_score(
            knn, X_train, y_train, scoring="accuracy", cv=10)
        # print(score)
        cv_scores.append(scores.mean())
    return cv_scores


def plt_knn_scores(cv_scores):
    plt.figure(figsize=(8, 6))
    plt.plot(cv_scores, "-o")
    plt.xlabel("knn-k value", fontsize=12)
    plt.ylabel("mean accuracy", fontsize=12)
    plt.grid()
    plt.title("KNN super-args of mean accuracy", fontsize=14)
    plt.show()

cv_scores = knn_model_train(X, target)
plt_knn_scores(cv_scores)

点击查看代码

# 最佳k值是13
X_train, X_test, y_train, y_test = train_test_split(X, target, test_size=0.3,
                                                    random_state=111, shuffle=True, stratify=target)

k = 13
knn_best = KNeighborsClassifier(k)
knn_best.fit(X_train, y_train)
print("泛化精度是,%.5f" % knn_best.score(X_test, y_test))