adaBoost-python实现

李航《统计学习方法》p140页例题

import numpy as np
import math
# 《统计学习方法》p140例子

def read_data():
    fp = open("dataSet\\adaBoostData.txt")
    sample = []; label = []
    for line in fp.readlines():
        arr = line.split()
        sample.append(int(arr[0]))
        label.append(int(arr[1]))
    fp.close()
    return sample, label


# 更新训练数据分布即权值weight, 更新基分类器的系数
def update(error, thro, dir):
    # 先更新及分类器系数
    alpha_m = 0.5 * math.log((1-error)/error)
    alpha.append(alpha_m)
    # 更新训练数据集的权值分布,先求规范化因子
    zm = 0
    for i in range(sample_n):
        gm = 1 if sample[i] < thro else -1
        weight[i] = weight[i] * math.exp(-alpha_m * label[i] * gm * dir)
        zm += weight[i]
    for i in range(sample_n):
        weight[i] /= zm

def cal_error(thro, dir):
    # 计算基分类器的分类结果
    error = 0
    for i in range(sample_n):
        error += weight[i] * ((sample[i] < thro) ^ (dir*label[i] > 0))
    return error


def get_score():
    score = 0.0
    for i in range(sample_n):
        score -= weight[i] * label[i]
    return score


# 训练基分类器，也就是训练阈值，不知道有没有简单的求阈值方法
def base_classifier():
    # 大于该阈值为-1， 小于为+1(暂时这么记录，符号以实际为准)
    score = np.zeros((1, sample_n+1))  # 记录不同阈值分类误差率
    score[0][0] = get_score()
    for i in range(1, sample_n+1):
        score[0][i] = score[0][i-1] + 2 * weight[i-1] * label[i-1]
    s_l = score[0].tolist()
    s_l_max = max(score[0])
    s_l_min = min(score[0])
    # 记录方向，dir=-1,小于阈值为-1，大于为+1
    if s_l_max <= math.fabs(s_l_min):
        dir = -1
        thro = s_l.index(s_l_min)-0.5
    else:
        dir = 1
        thro = s_l.index(s_l_max)-0.5
    return thro, dir


if __name__ == '__main__':
    sample, label = read_data()
    sample_n = np.shape(label)[0]
    weight = 0.1 * np.ones(np.shape(label))
    thro = []; alpha = []
    # 训练3个基分类器, 分类器系数存于alpha中
    for i in range(3):
        thro_tmp, dir_tmp = base_classifier()
        thro.append(thro_tmp)
        error = cal_error(thro_tmp, dir_tmp)
        update(error, thro_tmp, dir_tmp)
    print('thro:')
    print(thro)
    print('alpha:')
    print(alpha)

adaBoostData.txt

0 1
1 1
2 1
3 -1
4 -1
5 -1
6 1
7 1
8 1
9 -1