KNN分類算法

原理

數據集

http://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/wdbc.data

在使用時我直接放入到了txt中並且加上了標題欄，要不要都無所謂，但是程序需要稍作修改不然就會少一條數據。

ID	Diagnosis	radius_mean	texture_mean	perimeter_mean	area_mean	smoothness_mean	compactness_mean	concavity_mean	concave_mean	symmetry_mean	fractal_mean	radius_sd	texture_sd	perimeter_sd	area_sd	smoothness_sd	compactness_sd	concavity_sd	concave_sd	symmetry_sd	fractal_sd	radius_max	texture_max	perimeter_max	area_max	smoothness_max	compactness_max	concavity_max	concave_max	symmetry_max	fractal_max

程序

寫的不好，歡迎指正

#!/usr/bin/env python
# -*- coding: utf-8 -*-
# author：albert time:2018/11/28
import random


def read_dataset(address, val_list):  # 將txt中的數據讀到list中
    f = open(address)
    colum = f.readline()
    while len(colum) != 0:
        val_list.append(colum.split('\t'))
        colum = f.readline()
    return val_list


def split_dataset(dataset, percent):  # 輸入想分割的數據集和分割比例，返回percent比例的訓練集和剩下的訓練集
    dataset.remove(dataset[0])  # 去表頭
    n = int(len(dataset) * percent)
    test_data =[]
    while n >0:
        number = random.randint(0, len(dataset)-1)
        test_data.append(dataset[number])
        dataset.remove(dataset[number])
        n = n-1
    return dataset, test_data


def get_label(data):
    label = []
    for i in data:
        label.append(i[1])
    return label


def min_max_train(data):  # 默認接受到的是二維數組，這個似乎還是需要根據不同的數據集做一些改變而不能通用的
    n1 = len(data)    # 樣本數
    n2 = len(data[0]) # 屬性數
    max_val = [0, 0]
    min_val = [0, 0]
    for i in range(2, n2):
        max_val.append(float(data[0][i]))
        min_val.append(float(data[0][i]))
        for j in range(n1):
            if max_val[i] < float(data[j][i]):
                max_val[i] = float(data[j][i])
            if min_val[i] > float(data[j][i]):
                min_val[i] = float(data[j][i])
        for j in range(n1):
            data[j][i] = (float(data[j][i]) - min_val[i]) / (max_val[i] - min_val[i])
    return data, max_val, min_val     # 返回標準化的訓練集和訓練集中每個屬性的最大最小值

def min_max_test(data, max_val, min_val):
    n1 = len(data)  # 樣本數
    n2 = len(data[0])  # 屬性數
    for i in range(2, n2):
        for j in range(n1):
            data[j][i] = (float(data[j][i]) - min_val[i]) / (max_val[i] - min_val[i])
    return data # 返回標準化的測試集


def distance(train_data, test_data):
    n_train = len(train_data)
    n_attribute = len(train_data[0])
    di = []
    for i in range(n_train):
        x = 0
        for j in range(2, n_attribute):
            x += (test_data[j]-train_data[i][j]) ** 2
        x = x ** 0.5
        di.append([])
        di[i].append(x)
        di[i].append(i)
    return di #返回距離和標號


def sort_k(distance , k):
    temp = []
    for i in range(k):
        temp.append(distance[i])
    for i in range(k):
        for j in range(i, k):
            if temp[i][0] > temp[j][0]:
                temp[i], temp[j] = temp[j], temp[i]
    for i in range(k,len(distance)):
        for j in range(k):
            if distance[i][0] < temp[j][0]:
                temp[j] = distance[i]
                break
    return temp


def find_result(k_di, train_data):
    label = []
    M = 0
    B = 0
    result = ['M', 'B', 'same']
    for i in range(len(k_di)):
        label.append(train_data[k_di[i][1]][1])
    for i in label:
        if i == 'M':
            M += 1
        if i == 'B':
            B += 1
    print('THE PERCENT OF M IS %f' % (M / len(k_di)))
    print('THE PERCENT OF B IS %f' % (B / len(k_di)))
    if M > B:
        return result[0]
    elif M < B:
        return result[1]
    else:
        return result[2]

file_address = 'C:/Users/J/Desktop/wdbc.data.txt' # 放的text數據
val_list = []
read_dataset(file_address, val_list)
train_data, test_data = split_dataset(val_list, 0.2)
train_data, max_val, min_val = min_max_train(train_data)
test_data = min_max_test(test_data, max_val, min_val)
di = distance(train_data, test_data[0])
print(di)
k_di = sort_k(di, 25)
print(k_di)
result = find_result(k_di, train_data)
#  最後只是取一個測試集的來展示一下效果，可以根據需求再改
print(result)
print(test_data[0][1])