批量梯度下降:
随机梯度下降:
学习率不能太大。
当学习率增大的时候,迭代次数也需要同倍的变化。
#!/usr/bin/python
# -*- coding: UTF-8 -*-
# 文件名: batch_gradient_descent.py
import numpy as np
__author__ = 'yasaka'
X = 2 * np.random.rand(100, 1)
y = 4 + 3 * X + np.random.randn(100, 1)
X_b = np.c_[np.ones((100, 1)), X]
# print(X_b)
learning_rate = 0.01
n_iterations = 100000
m = 100
# 1,初始化theta,w0...wn
theta = np.random.randn(2, 1)
# 4,不会设置阈值,直接设置超参数,迭代次数,迭代次数到了,我们就认为收敛了
for iteration in range(n_iterations):
# 2,接着求梯度gradient
gradients = X_b.T.dot(X_b.dot(theta)-y)
# 3,应用公式调整theta值,theta_t + 1 = theta_t - grad * learning_rate
theta = theta - learning_rate * gradients
print(theta)