从零开始写代码AdaBoost算法的python实现

视频版见B站：Python实现AdaBoost算法-从零开始写代码_哔哩哔哩_bilibili

源文件、训练数据、说明图片下载：https://files.cnblogs.com/files/ljy1227476113/AdaBoost%E5%88%86%E7%B1%BB%E7%AE%97%E6%B3%95.7z

 

 

 

# author:会武术之白猫
# date:2021-11-11

import csv
import numpy as np
import random

def read_csv(csv_name):
    with open(csv_name) as f:
        reader = csv.reader(f)
        rows = [row for row in reader]

    data_x = []
    res_y = []
    for item in rows:
        if item[-1] == 'label':
            continue
        jtem = [float(t) for t in item[:-2]]
        data_x.append(jtem)
        res_y.append(int(item[-1]))
    # for item in data_x:
    #     print(item)
    # print(res_y)
    return data_x, res_y

# read_csv("iris_data.csv")

def train(x, y, w):
    min_loss = 1
    m = x.shape[1]
    res_k = None
    nums = 10
    for t in range(nums):
        k_random = []
        for i in range(m):
            k_random.append(random.uniform(0, 1))
        k_random = np.array(k_random)
        mis = (k_random*x).sum(axis = 1)
        mis = mis - min(mis)
        mis = mis / max(mis)
        res = []
        for item in mis:
            if item <= 0.33:
                res.append(0)
            elif item <= 0.66:
                res.append(1)
            else:
                res.append(2)
        res = np.array(res)
        miss = sum((res != y)*w)
        if miss < min_loss:
            min_loss = miss
            res_k = k_random
        # percent = sum(res == y)/n
        # print("正确率为{}%".format(percent*100))
    #print(min_loss)
    return min_loss, res_k

# x, y = read_csv("iris_data.csv")
# x = np.array(x)
# y = np.array(y)
# n = x.shape[0]
# M = 1
# w_m = np.array([1/n]*n)
# res = np.zeros(n)
# train(x, y, w_m)

def predict(x, res_k):
    mis = (res_k*x).sum(axis = 1)
    mis = mis - min(mis)
    mis = mis / max(mis)
    res = []
    for item in mis:
        if item <= 0.33:
            res.append(0)
        elif item <= 0.66:
            res.append(1)
        else:
            res.append(2)
    res = np.array(res)
    return res

def adaboost(csv_name):
    x, y = read_csv(csv_name)
    x = np.array(x)
    y = np.array(y)
    n = x.shape[0]
    M = 4
    w_m = np.array([1/n]*n)
    res = np.zeros(n)
    for m in range(M):
        e_m, res_k = train(x, y, w_m)
        a_m = 1/2 * np.log((1 - e_m)/e_m)
        y_m = predict(x, res_k)
        w_m = w_m * np.exp(-a_m*y*y_m)
        z_m = np.sum(w_m)
        w_m = w_m/z_m
        res += a_m*y_m
    res = res - min(res)
    res = res / max(res)
    result = []
    for item in res:
        if item <= 0.33:
            result.append(0)
        elif item <= 0.66:
            result.append(1)
        else:
            result.append(2)
    result = np.array(result)
    # print(result)
    percent = sum(result == y)/n
    print("正确率为{}%".format(percent*100))

csv_name = "iris_data.csv"
adaboost(csv_name)