python 10大算法之一 LinearRegression 笔记

时间:2023-01-22 14:49:03

简单的线性回归预测房价

#!/usr/bin/env python
# encoding: utf-8 """
@version:
@author: --*--.
@file: LinearRegression.py
@time: 2018/11/1 11:05
@desc:
""" # Required Packages
import matplotlib.pyplot as plt
import numpy as np
from sklearn import linear_model def get_data():
"""
生成随机的线性数据集
:return:
"""
x = 100 * np.random.rand(100, 1).astype(np.float32)
y = 2 * x + 10 # 直线
# y = 7 * x ** 5 + 3 * x + 10 # 曲线
y += 50 * np.random.rand(100, 1).astype(np.float32) return x, y # Function for Fitting our data to Linear model
def linear_model_main(X_parameters, Y_parameters, predict_value):
# Create linear regression object
regr = linear_model.LinearRegression()
regr.fit(X_parameters, Y_parameters,sample_weight=None) # 权重
predict_outcome = regr.predict(predict_value) predictions = {}
predictions['intercept'] = regr.intercept_
predictions['coefficient'] = regr.coef_
predictions['predicted_value'] = predict_outcome
return predictions # Function to show the resutls of linear fit model
def show_linear_line(X_parameters, Y_parameters, predictvalue):
# Create linear regression object
regr = linear_model.LinearRegression()
regr.fit(X_parameters, Y_parameters) fig = plt.figure()
ax1 = fig.add_subplot(111)
# 设置标题
ax1.set_title('Housing Forecast') ax1.scatter(X_parameters, Y_parameters, color='blue', marker='*')
ax1.plot(X_parameters, regr.predict(X_parameters), color='c', linewidth=1) # 画点
ax1.scatter(predictvalue, regr.predict(predictvalue), color='red') # 画水平虚线
plt.axvline(x=predictvalue, ls='dotted', color='y')
plt.axhline(y=regr.predict(predictvalue), ls='dotted', color='y') plt.xlabel('x:area')
plt.ylabel('y:price') plt.show() if __name__ == "__main__":
X, Y = get_data()
predictvalue = 90 # 面积
# 新版必须2维哦
predictvalue = np.array(predictvalue,dtype=np.int32).reshape(1, -1) result = linear_model_main(X, Y, predictvalue) print("截距-Intercept value ", result['intercept'])
print("回归系数-coefficient", result['coefficient'])
print("y-Predicted value: ", result['predicted_value'])
print("面积 %d 的价格预测为 %d" % (predictvalue, result['predicted_value'])) show_linear_line(X, Y, predictvalue)

输出结果为:

/usr/bin/python3.5 /home/think-hxr/PycharmProjects/MachineLearningAlgorithms/LinearRegression.py
Intercept value [ 38.77058411]
coefficient [[ 1.92119944]]
Predicted value: [[ 211.67853379]]
面积 90 的价格预测为 211

画图:

python 10大算法之一 LinearRegression 笔记