python——聚类

Cluster Analysis聚类分析

1. Introduction to Unsupervised Learning
2. Clustering
3. Similarity or Distance Calculation
4. Clustering as an Optimization Function
5. Types of Clustering Methods
6. Partitioning Clustering - KMeans & Meanshift
7. Hierarchial Clustering - Agglomerative
8. Density Based Clustering - DBSCAN
9. Measuring Performance of Clusters

1.无监督学习简介

2.聚类

3.相似度或距离计算

4.聚类作为优化函数

5.聚类方法的类型

6.分区聚类-KMeans和Meanshift

7.层次聚类-聚集

8.基于密度的群集-DBSCAN

9.衡量集群的绩效

10.比较所有聚类方法

Introduction to Unsupervised Learning无监督学习简介

• Unsupervised Learning is a type of Machine learning to draw inferences from unlabelled datasets.
• Model tries to find relationship between data.
• Most common unsupervised learning method is clustering which is used for exploratory data analysis to find hidden patterns or grouping in data
• 无监督学习是一种机器学习，可以从未标记的数据集中得出推论。
• 模型试图查找数据之间的关系。
• 最常见的无监督学习方法是聚类，用于探索性数据分析以发现隐藏模式或数据分组

Clustering

• A learning technique to group a set of objects in such a way that objects of same group are more similar to each other than from objects of other group.

• Applications of clustering are as follows

  • Automatically organizing the data
  • Labeling data
  • Understanding hidden structure of data
  • News Cloustering for grouping similar news together
  • Customer Segmentation
  • Suggest social groups

• 一种将一组对象进行分组的学习技术，使得同一组的对象比来自其他组的对象彼此更相似。

• 集群的应用如下:

  • 自动整理数据
  • 标签数据
  • 了解数据的隐藏结构
  • 新闻汇总，将相似的新闻分组在一起
  • 客户细分
  • 建议社交团体

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
from sklearn.datasets import make_blobs#make_blobs 产生多类数据集，对每个类的中心和标准差有很好的控制

• Generating natural cluster
• 生成自然簇

X,y = make_blobs(n_features=2, n_samples=1000, centers=3, cluster_std=1, random_state=3)
plt.scatter(X[:,0], X[:,1], s=5, alpha=.5)

Distance or Similarity Function距离或相似度函数

• Data belonging to same cluster are similar & data belonging to different cluster are different.
• We need mechanisms to measure similarity & differences between data.
• This can be achieved using any of the below techniques.

• Minkowiski breed of distance calculation:

• Manhatten (p=1), Euclidian (p=2)

• Cosine: Suited for text data

• 属于同一集群的数据相似，而属于不同集群的数据则不同。

• 我们需要一种机制来衡量数据之间的相似性和差异。

• 这可以使用以下任何一种技术来实现。

  • Minkowiski距离计算品种：
  

  • 曼哈顿（p = 1），欧几里得（p = 2）

  • 余弦：适用于文本数据

  

from sklearn.metrics.pairwise import euclidean_distances,cosine_distances,manhattan_distances
X = [[0, 1], [1, 1]]
print(euclidean_distances(X, X))
print(euclidean_distances(X, [[0,0]]))
print(euclidean_distances(X, [[0,0]]))
print(manhattan_distances(X,X))

[[0. 1.]
 [1. 0.]]
[[1.        ]
 [1.41421356]]
[[1.        ]
 [1.41421356]]
[[0. 1.]
 [1. 0.]]

Clustering as an Optimization Problem聚类是一个优化问题

• Maximize inter-cluster distances

• Minimize intra-cluster distances

• 最大化集群间距离

• 最小化集群内距离

Types of Clustering聚类的类型

• Partitioning methods
  • Partitions n data into k partitions
  • Initially, random partitions are created & gradually data is moved across different partitions.
  • It uses distance between points to optimize clusters.
  • KMeans & Meanshift are examples of Partitioning methods
• Hierarchical methods
  • These methods does hierarchical decomposition of datasets.
  • One approach is, assume each data as cluster & merge to create a bigger cluster
  • Another approach is start with one cluster & continue splitting
• Density-based methods
  • All above techniques are distance based & such methods can find only spherical clusters and not suited for clusters of other shapes.
  • Continue growing the cluster untill the density exceeds certain threashold.
• 分区方法
  • 将n个数据分区为k个分区
  • 最初，创建随机分区，然后逐渐在不同分区之间移动数据。
  • 它使用点之间的距离来优化聚类。
  • KMeans和Meanshift是分区方法的示例
• 分层方法
  • 这些方法对数据集进行分层分解。
  • 一种方法是，将每个数据假定为群集并合并以创建更大的群集
  • 另一种方法是从一个群集开始并继续拆分
• 基于密度的方法
  • 所有上述技术都是基于距离的，并且此类方法只能找到球形簇，而不适合其他形状的簇。
  • 继续生长群集，直到密度超过特定阈值。

Partitioning Method

KMeans

• Minimizing creteria : within-cluster-sum-of-squares.

• The centroids are chosen in such a way that it mi