机器学习之KNN

机器学习之KNN

约会问题为例，展示整个算法工作流程

源代码

注：这里使用的是python语言

import numpy as np
import operator

def read_data(file_path):
    fr = open(file_path)
    file_lines = fr.readlines()
    data_num = len(file_lines)
    dataset = np.zeros((data_num, 3))
    class_vec = []
    index = 0
    for line in file_lines:
        line_split = (line.strip()).split('\t')
        dataset[index, :] = line_split[:-1]
        class_vec.append(line_split[-1])
        index += 1
    return dataset, class_vec

def data_classify(dataset, class_vec):
    type1 = [[], [], []]
    type2 = [[], [], []]
    type3 = [[], [], []]
    labels = list(set(class_vec))
    
    for i in range(len(dataset)):
        if class_vec[i] == labels[0]:
            type1[0].append(dataset[i,0])
            type1[1].append(dataset[i, 1])
            type1[2].append(dataset[i, 2])
        elif class_vec[i] == labels[1]:
            type2[0].append(dataset[i,0])
            type2[1].append(dataset[i, 1])
            type2[2].append(dataset[i, 2])
        elif class_vec[i] == labels[2]:
            type3[0].append(dataset[i,0])
            type3[1].append(dataset[i, 1])
            type3[2].append(dataset[i, 2])

    return type1, type2, type3, labels

def view_data(dataset, class_vec):
    from matplotlib import pyplot as plt
    plt.rcParams['font.sans-serif'] = ['Simhei']
    plt.rcParams['axes.unicode_minus'] = False
    type1, type2, type3, labels = data_classify(dataset, class_vec)
    fig1 = plt.figure(figsize=(30, 10))

    axe1 = plt.subplot(1, 3, 1)
    axe1.scatter(type1[0], type1[1])
    axe1.scatter(type2[0], type2[1])
    axe1.scatter(type3[0], type3[1])
    plt.legend(labels)

    axe2 = plt.subplot(1, 3, 2)
    axe2.scatter(type1[0], type1[2])
    axe2.scatter(type2[0], type2[2])
    axe2.scatter(type3[0], type3[2])
    plt.legend(labels)

    axe3 = plt.subplot(1, 3, 3)
    axe3.scatter(type1[1], type1[2])
    axe3.scatter(type2[1], type2[2])
    axe3.scatter(type3[2], type3[2])
    plt.legend(labels)
    plt.show()

def split_dataset(dataset, class_vec):
    ratio = 0.8
    stop_index = int(dataset.shape[0] * ratio)
    train_dataset = dataset[: stop_index, :]
    test_dataset = dataset[stop_index:, :]

    train_class = class_vec[: stop_index]
    test_class = class_vec[stop_index :]

    return train_dataset, test_dataset, train_class, test_class

def data_normalize(train_dataset):
    maxv = train_dataset.max(0)
    minv = train_dataset.min(0)
    diff = maxv -minv
    # print(diff, type(diff))
    data_normal = (train_dataset -  np.tile(minv, (train_dataset.shape[0],1))) / np.tile(diff, (train_dataset.shape[0],1))
    # print(data_normal[:10,:])
    return data_normal, maxv, minv

def KNN(train_dataset, train_class, input_data, k):
    difference = np.sum((train_dataset - input_data) ** 2, 1)
    sorted_index = difference.argsort()
    class_count = {}
    for i in range(k):
        voted_label = train_class[sorted_index[i]]
        class_count[voted_label] = class_count.get(voted_label, 0) + 1

    sorted_class = sorted(class_count.items(), key = operator.itemgetter(1), reverse=True)
    return sorted_class[0][0]

def main():
    file_path = "datingTestSet.txt"
    dataset, class_vec = read_data(file_path)
    print('dataset = \n', dataset[:5, :])
    print(type(class_vec), 'class_vec = \n', class_vec[:20])

    view_data(dataset, class_vec)
    print('画图完成！')

    train_dataset, test_dataset, train_class, test_class = split_dataset(dataset, class_vec)
    train_data_normal, maxv, minv = data_normalize(train_dataset)
    error_num = 0
    for i in range(len(test_dataset)):
        data = test_dataset[i]
        data_normal = (data - minv) / (maxv - minv)
        label = KNN(train_data_normal, train_class, data_normal, 3)
        print('估计出的标签为%s, 实际的标签为：%s' % (label, test_class[i]))
        if str(label) != str(test_class[i]):
            error_num += 1
            print('*' * 20, '上面一行错误', '*' * 20)
    print('错误率：%.2f' % (float(error_num / len(test_dataset))))

if __name__ == "__main__":
    main()