西瓜书3.3logstic回归实现

|

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import mathdef sigmoid(x):return 1/(1 &＃43; math.e ** (-x))def cal(w1, w2, x1, x2, b):return sigmoid(w1*x1 &＃43; w2*x2 &＃43; b)data &＃61; pd.read_csv(&＃39;data3_0.csv&＃39;,encoding &＃61; "gbk")
X1 &＃61; data[u"密度"]
X2 &＃61; data[u"甜度"]
Label &＃61; data[u"好or坏"]
w1 &＃61; 0.0
w2 &＃61; 0.0
b &＃61; 0.0
learing_rate &＃61; 0.5
acc &＃61; 0.0
epoch &＃61; 0for epoch in range(1000):delta_w &＃61; np.zeros((2,1),dtype &＃61; float)delta_b &＃61; 0.0accurate &＃61; 0for i in range(len(X1)):x1 &＃61; X1[i]x2 &＃61; X2[i]label &＃61; Label[i]predict_y &＃61; cal(w1,w2,x1,x2,b)if (predict_y >&＃61; 0.5) &＃61;&＃61; label:accurate &＃61; accurate &＃43; 1delta_w[0] &＃43;&＃61; learing_rate*x1*(label - predict_y)delta_w[1] &＃43;&＃61; learing_rate*x2*(label - predict_y)delta_b &＃43;&＃61; learing_rate*(label - predict_y)w1 &＃43;&＃61; delta_w[0]w2 &＃43;&＃61; delta_w[1]b &＃43;&＃61; delta_b)acc &＃61; accurate / len(X1)
plt.scatter(X1,X2,c &＃61; Label)
plt.show()print("epoch %d&＃39;s accurary &＃61; %f%%" % (epoch,acc*100))
print("w1 &＃61; %f,w2 &＃61; %f,b &＃61; %f"% (w1,w2,b))

分析

最终输出:

epoch 999&＃39;s accurary &＃61; 70.588235%w1 &＃61; 3.156135,w2 &＃61; 12.541667,b &＃61; -4.431267


散点图:

可以看出此图并不便于进行二分类&＃xff0c;这也是逻辑回归在此数据集表现不佳的原因