逻辑回归的HHH可以看做是一个线性回归方程的结果经过一个sigmoid函数得到的结果(为正样本的概率),逻辑回归的假设函数如下:
hθ(x)=g(θTx)=11+e−θTx
h _ { \theta } ( x ) = g \left( \theta ^ { T } x \right) = \frac { 1 } { 1 + e ^ { - \theta ^ { T } x } }
hθ(x)=g(θTx)=1+e−θTx1
函数 hθ(x)h _ { \theta } ( x )hθ(x) 表示样本被预测为正例 111 的概率,我们很容易的得到样本被预测为正例和负例的概率如下:
P(y=1∣x; θ)=hθ(x)P(y=0∣x; θ)=1−hθ(x)
\begin{array} { l } P ( y = 1 | x ;\ \theta ) = h_{\theta} ( x ) \\ P ( y = 0 | x ; \ \theta ) = 1 - h _ { \theta } ( x ) \end{array}
P(y=1∣x; θ)=hθ(x)P(y=0∣x; θ)=1−hθ(x)
上式可以合并为一个式子:(预测结果的概率表示)
P(y∣x;θ)=(hθ(x))y(1−hθ(x))1−y
P ( y | x ; \theta ) = \left( h _ { \theta } ( x ) \right) ^ { y } \left( 1 - h _ { \theta } ( x ) \right) ^ { 1 - y }
P(y∣x;θ)=(hθ(x))y(1−hθ(x))1−y
我们对“预测结果的概率表示”取似然函数,取似然函数就是将模型对样本的概率预测值累乘起来。得到如下的似然函数:
L(θ)=∏i=1mP(y(i)∣x(i);θ)=∏i=1m(hθ(x(i)))y(i)(1−hθ(x(i)))1−y(i)
L ( \theta ) = \prod _ { i = 1 } ^ { m } P \left( y ^ { ( i ) } | x ^ { ( i ) } ; \theta \right) = \prod _ { i = 1 } ^ { m } \left( h _ { \theta } \left( x ^ { ( i ) } \right) \right) ^ { y ^ { ( i ) } } \left( 1 - h _ { \theta } \left( x ^ { ( i ) } \right) \right) ^ { 1 - y ^ { ( i ) } }
L(θ)=i=1∏mP(y(i)∣x(i);θ)=i=1∏m(hθ(x(i)))y(i)(1−hθ(x(i)))1−y(i)
由于该式比较麻烦涉及连乘法,所以我们对其去加对数操作得到对数似然函数:
上述利用的是最大似然估计原理:极大似然估计就是利用已知的样本分布,找到最有可能(即最大概率)导致这种分布的参数值;或者说什么样的参数才能使我们观测到目前这组数据的概率最大。
l(θ)=logL(θ)=∑i=1m(y(i)loghθ(x(i))+(1−y(i))log(1−hθ(x(i))))
l ( \theta ) = \log L ( \theta ) = \sum _ { i = 1 } ^ { m } \left( y ^ { ( i ) } \log h _ { \theta } \left( x ^ { ( i ) } \right) + \left( 1 - y ^ { ( i ) } \right) \log \left( 1 - h _ { \theta } \left( x ^ { ( i ) } \right) \right) \right)
l(θ)=logL(θ)=i=1∑m(y(i)loghθ(x(i))+(1−y(i))log(1−hθ(x(i))))
当似然函数求得最大值时,模型参数能够最大可能的满足当前的样本,求最大值使用梯度向上法,我们可以对似然函数加个负号,通过求等价问题的最小值来求原问题的最大值,这样我们就可以使用极大似然估计法。(注意这里还多加了个1m\frac { 1 } { m }m1)
J(θ)=−1ml(θ)
J ( \theta ) = - \frac { 1 } { m } l ( \theta )
J(θ)=−m1l(θ)
这样我们就能得到损失函数的最终形式:
J(θ)=−1m∑i=1m(y(i)loghθ(x(i))+(1−y(i))log(1−hθ(x(i))))
J ( \theta ) = - \frac { 1 } { m } \sum _ { i = 1 } ^ { m } \left( y ^ { ( i ) } \log h _ { \theta } \left( x ^ { ( i ) } \right) + \left( 1 - y ^ { ( i ) } \right) \log \left( 1 - h _ { \theta } \left( x ^ { ( i ) } \right) \right) \right)
J(θ)=−m1i=1∑m(y(i)loghθ(x(i))+(1−y(i))log(1−hθ(x(i))))
即等价于:
cost(hθ(x), y)={−log(hθ(x)) if y=1−log(1−hθ(x)) if y=0
\operatorname { cost } ( h _ { \theta } ( x ) ,\ y ) = \left\{ \begin{array} { l l } - \log \left( h _ { \theta } ( x ) \right) & \text { if } y = 1 \\ - \log \left( 1 - h _ { \theta } ( x ) \right) & \text { if } y = 0 \end{array} \right.
cost(hθ(x), y)={−log(hθ(x))−log(1−hθ(x)) if y=1 if y=0
我们下图为推导式,面试推导的时候可以不写下标(假设我们使用随机梯度下降法),这样可以使推导式更简洁。
求梯度:
这里需要提一下的是,sigmoid函数有如下性质,在上述推导的第三行中可以看到:
S′(x)=e−x(1+e−x)2=S(x)(1−S(x))
S ^ { \prime } ( x ) = \frac { e ^ { - x } } { \left( 1 + e ^ { - x } \right) ^ { 2 } } = S ( x ) ( 1 - S ( x ) )
S′(x)=(1+e−x)2e−x=S(x)(1−S(x))
θ更新式:α 为学习率
θj:=θj−α1m∑i=1m(hθ(x(i))−y(i))xj(i)
\theta _ { j } : = \theta _ { j } - \alpha \frac { 1 } { m } \sum _ { i = 1 } ^ { m } \left( h _ { \theta } \left( x ^ { ( i ) } \right) - y ^ { ( i ) } \right) x _ { j } ^ { ( i ) }
θj:=θj−αm1i=1∑m(hθ(x(i))−y(i))xj(i)
总结:LR在确定了模型的形式后,通过最大似然估计法来实现最小散度从而求出模型参数。
代码实现向量化:向量化是使用矩阵计算来代替for
循环,以简化计算过程,提高效率。
# -*- coding: utf-8 -*-
from numpy import *
from matplotlib import pyplot as plt
def plot_best_fit(wei, data_set, label):
weights = wei
data_set = array(data_set)
n = shape(data_set)[0]
xcourd1 = []; ycourd1 = []
xcourd2 = []; ycourd2 = []
for i in range(n):
if int(label[i]) == 1:
xcourd1.append(data_set[i, 1]); ycourd1.append(data_set[i, 2])
else:
xcourd2.append(data_set[i, 1]); ycourd2.append(data_set[i, 2])
fig = plt.figure()
ax = fig.add_subplot(111)
ax.scatter(xcourd1, ycourd1, s=30, c='red', marker='s')
ax.scatter(xcourd2, ycourd2, s=30, c='green')
x = arange(-3.0, 3.0, 0.1)
y = (-weights[0] - weights[1]*x)/weights[2]
ax.plot(x, y)
plt.xlabel('X1'); plt.ylabel('X2')
plt.show()
def load_data():
data_set = []
label = []
fr = open('./text.txt')
for line in fr.readlines():
line = line.strip().split()
data_set.append([1.0, float(line[0]), float(line[1])])
label.append(int(line[2]))
return data_set, label
def sigmoid(x):
return 1.0 / (1 + exp(-x))
# 梯度下降算法 GD
def train(data_set, label):
data_matrix = mat(data_set)
label = mat(label).transpose()
m, n = shape(data_matrix)
alpha = 0.001
max_cycles = 500
weights = ones((n, 1))
for k in range(max_cycles):
h = sigmoid(data_matrix*weights)
error = h - label
weights = weights - alpha * data_matrix.transpose() * error
return weights
# on line to study SGD
def stoc_grad_descent(data_set, label):
m, n = shape(data_set)
alpha = 0.01
weights = ones(n)
for i in range(m):
h = sigmoid(sum(data_set[i]*weights))
error = h - label[i]
weights = weights - alpha * error * data_set[i]
return weights
# on line to study prove
def prove_grad_ascent(data_set, label, num_iter=450):
m, n = shape(data_set)
weights = ones(n)
for j in range(num_iter):
data_index = range(m)
for i in range(m):
alpha = 4/(1.0+j+i)+0.01 # prevent swings
# choose a random value to prevent periodic swings
rand_index = int(random.uniform(0, len(data_index)))
h = sigmoid(sum(data_set[rand_index]*weights))
error = label[rand_index] - h
weights = weights + alpha * error * data_set[rand_index]
del data_index[rand_index]
return weights
if __name__ == "__main__":
data_set, label = load_data()
#print label
#weights = train(array(data_set), label)
#weights = stoc_grad_ascent(array(data_set), label)
weights = prove_grad_ascent(array(data_set), label)
plot_best_fit(weights, data_set, label)
References
https://blog.csdn.net/dpengwang/article/details/86746233
https://www.jianshu.com/p/471b2fd570a3