homework

k-means.py

from math import *
import random
import numpy as np
import collections

class Kmeans:
    def __init__(self, cluster_num):
        self.cluster_num = cluster_num
        self.counter = 0
        self.error = 0.
        
    # データの読み込み
    def read_data(self, data_path):
        with open(data_path) as file:
            self.data = np.array(list(map(lambda x: list(map(int, x.split(','))), file.read().split('\n'))))

        self.label = [0 for i in range(len(self.data))]
        self.centroid = np.array([self.data[random.randint(0, len(self.data))] for i in range(self.cluster_num)])

    # ラベリング処理
    def labeling(self):
        self.error = 0.
        for i in range(len(self.data)):
            dist = []
            for c in self.centroid:
                dist.append(np.linalg.norm(self.data[i]-c))
            self.label[i] = dist.index(min(dist))
            self.error += min(dist)

    # 重心の更新
    def update_centroid(self):
        self.centroid = np.array([[0. for i in range(4)] for i in range(self.cluster_num)])
        cluster_counter = collections.Counter(self.label)
        for i in range(len(self.data)):
            self.centroid[self.label[i]] += (self.data[i] / cluster_counter[self.label[i]])

    # 学習
    def fit(self, epoch=10):
        for i in range(epoch):
            print("--------epoch {}---------".format(self.counter))
            print("centroid:")
            print(self.centroid)
            print("label:")
            print(self.label)            
            print("error:")
            print(self.error)

            self.labeling()
            self.update_centroid()
            self.counter += 1

if __name__ == '__main__':
    clustering = Kmeans(5)
    clustering.read_data('data.csv')
    clustering.fit(epoch=3)