Linear Regression

时间:2020-04-13 13:56:19   收藏:0   阅读:57

原创转载请注明出处:https://www.cnblogs.com/agilestyle/p/12690755.html

 

准备数据

import matplotlib.pyplot as plt
from sklearn.datasets import make_regression
from sklearn.linear_model import LinearRegression
from sklearn.linear_model import SGDRegressor
from sklearn.metrics import mean_squared_error

X, y = make_regression(n_samples=500, n_features=1, noise=20, random_state=0)

# (500, 1)
X.shape

# (500,)
y.shape

plt.scatter(X, y)

技术分享图片

 

建模训练

lr = LinearRegression()
y = y.reshape(-1, 1)

# LinearRegression(copy_X=True, fit_intercept=True, n_jobs=None, normalize=False)
lr.fit(X, y)

 

评价模型

# 查看模型的截距 array([-1.53821792])
lr.intercept_

# 查看模型的斜率 array([[45.2879203]])
lr.coef_

# 0.8458565184565707
lr.score(X, y)

lr_mse = mean_squared_error(y, lr.predict(X))
# 均方误差 372.3837648686677
lr_mse

plt.scatter(X, y)
plt.plot(X, lr.predict(X), r)

技术分享图片

 

使用随机梯度下降求解参数

sgd = SGDRegressor(eta0=0.01, max_iter=100)  # eta0: 初始学习率,max_iter: 最大迭代次数
sgd.fit(X, y.ravel())
sgd.score(X, y)

Note:

梯度下降的两个重要因素

 

Reference

scikit-learn Cookbook Second Edition

https://scikit-learn.org/stable/modules/generated/sklearn.metrics.mean_squared_error.html

https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.SGDRegressor.html

https://docs.scipy.org/doc/numpy/reference/generated/numpy.ravel.html

https://developers.google.com/machine-learning/crash-course/reducing-loss/gradient-descent

https://developers.google.com/machine-learning/crash-course/reducing-loss/learning-rate

 

原文:https://www.cnblogs.com/agilestyle/p/12690755.html

评论(0
© 2014 bubuko.com 版权所有 - 联系我们:wmxa8@hotmail.com
打开技术之扣,分享程序人生!