学习算法实现

一些学习算法实现练习

PLA 感知机学习算法

参考资料：

• 写给大家看的机器学习书（第三篇）

  # -*- coding: utf-8 -*-
  import numpy as np
  
  # 将列表转为列向量
  def vector(list):
  return np.mat(list).transpose()
  
  # 向x向量添加x0=1 将data格式化为(x，y)的格式
  def pre_process_data(list):
  return [(vector([1, l[0], l[1]]), l[2]) for l in list]
  
  # 符号函数 
  sign = lambda x:1 if x > 0 else -1
  
  # 以w为参数 遍历所有训练数据 判断训练数据中是否有点被切分错误
  def go_through_all_training_data(training_data, w):
  status = 'YES'
  for (x, y) in training_data:
      if sign((w.transpose() * x).tolist()[0][0]) != sign(y):
          status = 'NO'
          return (status, x, y)
  return (status, None, None)
  
  # 感知机学习算法 主函数
  def PLA(training_data):
  w = np.mat([1,2127,205]).transpose()  # Step 1: 向量w初始化
  while True:
      (status, x, y) = go_through_all_training_data(training_data, w)
      if status == 'YES':  # Step 2: 切分正确，学习完成
          return w
      else:
          w = w + y*x   # Step 3: 修正w
  
  if __name__ == '__main__':
  # 训练数据
  training_data = [
  [10, 300, -1],
  [15, 377, -1],
  [50, 137, 1],
  [65, 92 , 1],
  [45, 528, -1],
  [61, 542, 1],
  [26, 394, -1],
  [37, 703, -1],
  [39, 244, 1],
  [41, 398, 1],
  [53, 495, 1],
  [32, 119, 1],
  [24, 577, -1],
  [56, 412, 1]
  ]
  
  formated_training_data = pre_process_data(training_data)
  
  w = PLA(formated_training_data)
  
  print(w)