GBRT-Regression-Demo(超参数的比较）

import matplotlib.pyplot as plt
import numpy as np
from sklearn import ensemble,datasets,model_selection

def load_data_regression():
    diabetes = datasets.load_diabetes()
    return model_selection.train_test_split(diabetes.data,diabetes.target,test_size=0.3,random_state = 0)  

def test_GradientBoostingRegression(*data):
    X_train,X_test,Y_train,Y_test = data
    regr = ensemble.GradientBoostingRegressor()
    regr.fit(X_train,Y_train)
    print("Training score:%f"%regr.score(X_train,Y_train))
    print("Testing score:%f"%regr.score(X_test,Y_test))

X_train,X_test,Y_train,Y_test = load_data_regression()
test_GradientBoostingRegression(X_train,X_test,Y_train,Y_test)

Training score:0.876188
Testing score:0.275905

考察个体回归树的数量对GBRT的影响

def test_GradientBoostingRegression_estimators_num(*data):
    X_train,X_test,Y_train,Y_test = data
    nums = np.arange(1,100,step=2)
    fig = plt.figure()
    ax = fig.add_subplot(1,1,1)
    train_scores = []
    test_scores = []
    for i,num in enumerate(nums):
        clf = ensemble.GradientBoostingRegressor(n_estimators=num)
        clf.fit(X_train,Y_train)
        train_scores.append(clf.score(X_train,Y_train))
        test_scores.append(clf.score(X_test,Y_test))
    ax.plot(nums,train_scores,label="training score")
    ax.plot(nums,test_scores,label="testing score")
    ax.set_xlabel("estimators num")
    ax.set_ylabel("score")
    ax.legend(loc="lower right")
    ax.set_title("GradientBoostRegression")
plt.show()

X_train,X_test,Y_train,Y_test = load_data_regression()
test_GradientBoostingRegression_estimators_num(X_train,X_test,Y_train,Y_test)
output_6_0.png
运行结果分析：
  1. 随着个体回归树数量的增长，GBRT的性能对与训练集一直在提高。
  2. 但是对于测试集测试得分先快速上升后基本缓慢下降。

考察个体回归树的最大深度对于集成回归器预测性能的影响

def test_GradientBoostingRegression_max_depth(*data):
    X_train,X_test,Y_train,Y_test = data
    maxdepths = np.arange(1,20+1)
    fig = plt.figure()
    ax = fig.add_subplot(1,1,1)
    train_scores = []
    test_scores = []
    for i,maxdepth in enumerate(maxdepths):
        clf = ensemble.GradientBoostingRegressor(max_depth=maxdepth,max_leaf_nodes=None)
        clf.fit(X_train,Y_train)
        train_scores.append(clf.score(X_train,Y_train))
        test_scores.append(clf.score(X_test,Y_test))
    ax.plot(maxdepths,train_scores,label="Training score")
    ax.plot(maxdepths,test_scores,label="Testing score")
    ax.set_xlabel("maxdepth num")
    ax.set_ylabel("score")
    ax.legend(loc="lower right")
    ax.set_ylim(-1,1.05)
    ax.set_title("GradientBoostingRegression")
    plt.show()

X_train,X_test,Y_train,Y_test = load_data_regression()
test_GradientBoostingRegression_max_depth(X_train,X_test,Y_train,Y_test)
output_10_0.png
运行结果分析：
  • 随着个体回归树的最大深度增长，对训练集的拟合越来越好，但是对测试集的拟合越来越差，产生了过拟合现象。

考察学习率对于GBRT的预测性能的影响

def test_GradientBoostingRegression_learning_rate(*data):
    X_train,X_test,Y_train,Y_test = data
    learing_rates = np.linspace(0.01,1.0)
    fig = plt.figure()
    ax = fig.add_subplot(1,1,1)
    train_scores = []
    test_scores = []
    for i,learing_rate in enumerate(learing_rates):
        clf = ensemble.GradientBoostingRegressor(learning_rate=learing_rate)
        clf.fit(X_train,Y_train)
        train_scores.append(clf.score(X_train,Y_train))
        test_scores.append(clf.score(X_test,Y_test))
    ax.plot(learing_rates,train_scores,label="Training score")
    ax.plot(learing_rates,test_scores,label="Testing score")
    ax.set_xlabel("learing_rate")
    ax.set_ylabel("score")
    ax.legend(loc="best")
    ax.set_ylim(-1,1.05)
    ax.set_title("GradientBoostingRegression_learning_rate")
plt.show()

X_train,X_test,Y_train,Y_test = load_data_regression()
test_GradientBoostingRegression_learning_rate(X_train,X_test,Y_train,Y_test)
output_14_0.png
运行结果分析：
  • GBRT对预测集的预测得分随着学习率的增长而一直降低

考察subsample的影响

def test_GradientBoostingRegression_subsample(*data):
    X_train,X_test,Y_train,Y_test = data
    subsamples = np.linspace(0.01,1.0)
    fig = plt.figure()
    ax = fig.add_subplot(1,1,1)
    train_scores = []
    test_scores = []
    for i,subsample in enumerate(subsamples):
        clf = ensemble.GradientBoostingRegressor(subsample=subsample,n_estimators=100)
        clf.fit(X_train,Y_train)
        train_scores.append(clf.score(X_train,Y_train))
        test_scores.append(clf.score(X_test,Y_test))
    ax.plot(subsamples,train_scores,label="Training score")
    ax.plot(subsamples,test_scores,label="Testing score")
    ax.set_xlabel("subsample")
    ax.set_ylabel("score")
    ax.legend(loc="best")
    ax.set_ylim(-1,1.05)
    ax.set_title("GradientBoostingRegression_subsample")
plt.show()

X_train,X_test,Y_train,Y_test = load_data_regression()
test_GradientBoostingRegression_subsample(X_train,X_test,Y_train,Y_test)
output_18_0.png
运行结果分析：
  • 本问题中，subsample对GBRT预测的影响不大，主要对GBRT的训练拟合能力起作用。

考察max_features参数的影响

def test_GradientBoostingRegression_max_features(*data):
    X_train,X_test,Y_train,Y_test = data
    max_features = np.linspace(0.01,1.0)
    fig = plt.figure()
    ax = fig.add_subplot(1,1,1)
    train_scores = []
    test_scores = []
    for i,features in enumerate(max_features):
        clf = ensemble.GradientBoostingRegressor(max_features=features,n_estimators=100)
        clf.fit(X_train,Y_train)
        train_scores.append(clf.score(X_train,Y_train))
        test_scores.append(clf.score(X_test,Y_test))
    ax.plot(max_features,train_scores,label="Training score")
    ax.plot(max_features,test_scores,label="Testing score")
    ax.set_xlabel("max_features")
    ax.set_ylabel("score")
    ax.legend(loc="best")
    ax.set_ylim(0,1.05)
    ax.set_title("GradientBoostingRegression_max_features")
plt.show()

X_train,X_test,Y_train,Y_test = load_data_regression()
test_GradientBoostingRegression_max_features(X_train,X_test,Y_train,Y_test)
output_22_0.png
运行结果分析：
  • GBRT对于特征集合的选取不是很敏感

考察损失函数的影响

def test_GradientBoostingRegression_loss(*data):
    X_train,X_test,Y_train,Y_test = data
    fig = plt.figure(figsize=(15,15))
    nums = np.arange(1,200,step=2)
    ###ls:平方损失函数
    ###lad:绝对值损失函数
    ###huber:上述两者的加权结合
    losses = ["ls","lad","huber"]
    ax = fig.add_subplot(2,1,1)
    alphas = np.linspace(0.01,1.0,endpoint=False,num=5)
    for alpha in alphas:
        test_scores = []
        train_scores = []
        for num in nums:
            regr = ensemble.GradientBoostingRegressor(n_estimators=num,loss="huber",alpha=alpha)
            regr.fit(X_train,Y_train)
            train_scores.append(regr.score(X_train,Y_train))
            test_scores.append(regr.score(X_test,Y_test))
        ax.plot(nums,train_scores,label="Training score:alpha=%f"%alpha)
        ax.plot(nums,test_scores,label="Testing score:alpha=%f"%alpha)
    ax.set_xlabel("estimator num")
    ax.set_ylabel("score")
    ax.legend(loc="best",framealpha=0.4)
    ax.set_ylim(0,1.05)
    ax.set_title("loss=huber")
    plt.suptitle("GradientBoostRegressor")
    ######绘制ls和lad#######    
    ax = fig.add_subplot(2,1,2)
    for loss in ["ls","lad"]:
        test_scores = []
        train_scores = []
        for num in nums:
            regr = ensemble.GradientBoostingRegressor(n_estimators=num,loss=loss)
            regr.fit(X_train,Y_train)
            train_scores.append(regr.score(X_train,Y_train))
            test_scores.append(regr.score(X_test,Y_test))
        ax.plot(nums,train_scores,label="Training score:loss=%s"%loss)
        ax.plot(nums,test_scores,label="Testing score:loss=%s"%loss)
        ax.set_xlabel("estimator num")
        ax.set_ylabel("score")
        ax.legend(loc="best",framealpha=0.4)
        ax.set_ylim(0,1.05)
        ax.set_title("loss=ls,lad")
        plt.suptitle("GradientBoostRegressor")
    plt.show()
output.png
最后编辑于 ：
©著作权归作者所有,转载或内容合作请联系作者
平台声明：文章内容（如有图片或视频亦包括在内）由作者上传并发布，文章内容仅代表作者本人观点，简书系信息发布平台，仅提供信息存储服务。

推荐阅读更多精彩内容