[Machine Learning From Scratch]-unsupervised_learning-partitioning_around_medoids

  
from __future__ import print_function, division
import numpy as np
from mlfromscratch.utils import normalize, euclidean_distance, Plot
from mlfromscratch.unsupervised_learning import PCA


class PAM():
    """A simple clustering method that forms k clusters by first assigning
    samples to the closest medoids, and then swapping medoids with non-medoid
    samples if the total distance (cost) between the cluster members and their medoid
    is smaller than prevoisly.
    Parameters:
    -----------
    k: int
        The number of clusters the algorithm will form.
    """
    def __init__(self, k=2):
        self.k = k

    def _init_random_medoids(self, X):
        """ Initialize the medoids as random samples """
        n_samples, n_features = np.shape(X)
        medoids = np.zeros((self.k, n_features))
        for i in range(self.k):
            medoid = X[np.random.choice(range(n_samples))]
            medoids[i] = medoid
        return medoids

    def _closest_medoid(self, sample, medoids):
        """ Return the index of the closest medoid to the sample """
        closest_i = None
        closest_distance = float("inf")
        for i, medoid in enumerate(medoids):
            distance = euclidean_distance(sample, medoid)
            if distance < closest_distance:
                closest_i = i
                closest_distance = distance
        return closest_i

    def _create_clusters(self, X, medoids):
        """ Assign the samples to the closest medoids to create clusters """
        clusters = [[] for _ in range(self.k)]
        for sample_i, sample in enumerate(X):
            medoid_i = self._closest_medoid(sample, medoids)
            clusters[medoid_i].append(sample_i)
        return clusters

    def _calculate_cost(self, X, clusters, medoids):
        """ Calculate the cost (total distance between samples and their medoids) """
        cost = 0
        # For each cluster
        for i, cluster in enumerate(clusters):
            medoid = medoids[i]
            for sample_i in cluster:
                # Add distance between sample and medoid as cost
                cost += euclidean_distance(X[sample_i], medoid)
        return cost

    def _get_non_medoids(self, X, medoids):
        """ Returns a list of all samples that are not currently medoids """
        non_medoids = []
        for sample in X:
            if not sample in medoids:
                non_medoids.append(sample)
        return non_medoids

    def _get_cluster_labels(self, clusters, X):
        """ Classify samples as the index of their clusters """
        # One prediction for each sample
        y_pred = np.zeros(np.shape(X)[0])
        for cluster_i in range(len(clusters)):
            cluster = clusters[cluster_i]
            for sample_i in cluster:
                y_pred[sample_i] = cluster_i
        return y_pred

    def predict(self, X):
        """ Do Partitioning Around Medoids and return the cluster labels """
        # Initialize medoids randomly
        medoids = self._init_random_medoids(X)
        # Assign samples to closest medoids
        clusters = self._create_clusters(X, medoids)

        # Calculate the initial cost (total distance between samples and
        # corresponding medoids)
        cost = self._calculate_cost(X, clusters, medoids)

        # Iterate until we no longer have a cheaper cost
        while True:
            best_medoids = medoids
            lowest_cost = cost
            for medoid in medoids:
                # Get all non-medoid samples
                non_medoids = self._get_non_medoids(X, medoids)
                # Calculate the cost when swapping medoid and samples
                for sample in non_medoids:
                    # Swap sample with the medoid
                    new_medoids = medoids.copy()
                    new_medoids[medoids == medoid] = sample
                    # Assign samples to new medoids
                    new_clusters = self._create_clusters(X, new_medoids)
                    # Calculate the cost with the new set of medoids
                    new_cost = self._calculate_cost(
                        X, new_clusters, new_medoids)
                    # If the swap gives us a lower cost we save the medoids and cost
                    if new_cost < lowest_cost:
                        lowest_cost = new_cost
                        best_medoids = new_medoids
            # If there was a swap that resultet in a lower cost we save the
            # resulting medoids from the best swap and the new cost 
            if lowest_cost < cost:
                cost = lowest_cost
                medoids = best_medoids 
            # Else finished
            else:
                break

        final_clusters = self._create_clusters(X, medoids)
        # Return the samples cluster indices as labels
        return self._get_cluster_labels(final_clusters, X)
?著作權(quán)歸作者所有,轉(zhuǎn)載或內(nèi)容合作請(qǐng)聯(lián)系作者
  • 序言:七十年代末寞钥,一起剝皮案震驚了整個(gè)濱河市慌申,隨后出現(xiàn)的幾起案子,更是在濱河造成了極大的恐慌理郑,老刑警劉巖蹄溉,帶你破解...
    沈念sama閱讀 222,865評(píng)論 6 518
  • 序言:濱河連續(xù)發(fā)生了三起死亡事件,死亡現(xiàn)場(chǎng)離奇詭異您炉,居然都是意外死亡柒爵,警方通過(guò)查閱死者的電腦和手機(jī),發(fā)現(xiàn)死者居然都...
    沈念sama閱讀 95,296評(píng)論 3 399
  • 文/潘曉璐 我一進(jìn)店門(mén)赚爵,熙熙樓的掌柜王于貴愁眉苦臉地迎上來(lái)棉胀,“玉大人,你說(shuō)我怎么就攤上這事冀膝⊙渖荩” “怎么了?”我有些...
    開(kāi)封第一講書(shū)人閱讀 169,631評(píng)論 0 364
  • 文/不壞的土叔 我叫張陵窝剖,是天一觀的道長(zhǎng)麻掸。 經(jīng)常有香客問(wèn)我,道長(zhǎng)赐纱,這世上最難降的妖魔是什么脊奋? 我笑而不...
    開(kāi)封第一講書(shū)人閱讀 60,199評(píng)論 1 300
  • 正文 為了忘掉前任熬北,我火速辦了婚禮,結(jié)果婚禮上诚隙,老公的妹妹穿的比我還像新娘讶隐。我一直安慰自己,他們只是感情好最楷,可當(dāng)我...
    茶點(diǎn)故事閱讀 69,196評(píng)論 6 398
  • 文/花漫 我一把揭開(kāi)白布整份。 她就那樣靜靜地躺著,像睡著了一般籽孙。 火紅的嫁衣襯著肌膚如雪烈评。 梳的紋絲不亂的頭發(fā)上,一...
    開(kāi)封第一講書(shū)人閱讀 52,793評(píng)論 1 314
  • 那天犯建,我揣著相機(jī)與錄音讲冠,去河邊找鬼。 笑死适瓦,一個(gè)胖子當(dāng)著我的面吹牛竿开,可吹牛的內(nèi)容都是我干的。 我是一名探鬼主播玻熙,決...
    沈念sama閱讀 41,221評(píng)論 3 423
  • 文/蒼蘭香墨 我猛地睜開(kāi)眼否彩,長(zhǎng)吁一口氣:“原來(lái)是場(chǎng)噩夢(mèng)啊……” “哼!你這毒婦竟也來(lái)了嗦随?” 一聲冷哼從身側(cè)響起列荔,我...
    開(kāi)封第一講書(shū)人閱讀 40,174評(píng)論 0 277
  • 序言:老撾萬(wàn)榮一對(duì)情侶失蹤,失蹤者是張志新(化名)和其女友劉穎枚尼,沒(méi)想到半個(gè)月后贴浙,有當(dāng)?shù)厝嗽跇?shù)林里發(fā)現(xiàn)了一具尸體,經(jīng)...
    沈念sama閱讀 46,699評(píng)論 1 320
  • 正文 獨(dú)居荒郊野嶺守林人離奇死亡署恍,尸身上長(zhǎng)有42處帶血的膿包…… 初始之章·張勛 以下內(nèi)容為張勛視角 年9月15日...
    茶點(diǎn)故事閱讀 38,770評(píng)論 3 343
  • 正文 我和宋清朗相戀三年崎溃,在試婚紗的時(shí)候發(fā)現(xiàn)自己被綠了。 大學(xué)時(shí)的朋友給我發(fā)了我未婚夫和他白月光在一起吃飯的照片盯质。...
    茶點(diǎn)故事閱讀 40,918評(píng)論 1 353
  • 序言:一個(gè)原本活蹦亂跳的男人離奇死亡袁串,死狀恐怖,靈堂內(nèi)的尸體忽然破棺而出呼巷,到底是詐尸還是另有隱情般婆,我是刑警寧澤,帶...
    沈念sama閱讀 36,573評(píng)論 5 351
  • 正文 年R本政府宣布朵逝,位于F島的核電站,受9級(jí)特大地震影響乡范,放射性物質(zhì)發(fā)生泄漏配名。R本人自食惡果不足惜啤咽,卻給世界環(huán)境...
    茶點(diǎn)故事閱讀 42,255評(píng)論 3 336
  • 文/蒙蒙 一、第九天 我趴在偏房一處隱蔽的房頂上張望渠脉。 院中可真熱鬧宇整,春花似錦、人聲如沸芋膘。這莊子的主人今日做“春日...
    開(kāi)封第一講書(shū)人閱讀 32,749評(píng)論 0 25
  • 文/蒼蘭香墨 我抬頭看了看天上的太陽(yáng)为朋。三九已至臂拓,卻和暖如春,著一層夾襖步出監(jiān)牢的瞬間习寸,已是汗流浹背胶惰。 一陣腳步聲響...
    開(kāi)封第一講書(shū)人閱讀 33,862評(píng)論 1 274
  • 我被黑心中介騙來(lái)泰國(guó)打工, 沒(méi)想到剛下飛機(jī)就差點(diǎn)兒被人妖公主榨干…… 1. 我叫王不留霞溪,地道東北人孵滞。 一個(gè)月前我還...
    沈念sama閱讀 49,364評(píng)論 3 379
  • 正文 我出身青樓,卻偏偏與公主長(zhǎng)得像鸯匹,于是被迫代替她去往敵國(guó)和親坊饶。 傳聞我的和親對(duì)象是個(gè)殘疾皇子,可洞房花燭夜當(dāng)晚...
    茶點(diǎn)故事閱讀 45,926評(píng)論 2 361

推薦閱讀更多精彩內(nèi)容