0 2022/7/19

1 clustering algorithm

• Clustering method based on partition , Such as K-means
  K-Means The algorithm is a typical clustering algorithm based on partition , It is also an unsupervised learning algorithm .K-Means The idea of the algorithm is very simple , For a given sample set , Euclidean distance is used to measure the similarity between data objects , Similarity is inversely proportional to the distance between data objects , The greater the similarity , The smaller the distance .
• Hierarchical clustering method , Such as CURE
  It is a condensation algorithm (AGNES). The algorithm first considers each data point as a class , Then merge the nearest class until the number of classes is the number required . But and AGNES The difference between algorithms is ： Cancel using all points or center points + Distance to represent a class , Instead, extract a fixed number from each class 、 Well distributed points are representative points of this kind _, And put these representative points ( commonly 10 individual ) Multiply by an appropriate contraction factor ( General Settings 0.2~0.7 Between ), Make them closer to the center of the class . The shrinkage properties of representative points can be adjusted to match those non spherical scenes , And the use of shrinkage factor can reduce the impact of noise on clustering .
• Grid based clustering method , Such as STING
  STING(Statistical Information Grid-based method) It is a grid based multi-resolution clustering technology , It divides the spatial area into rectangular units . For different levels of resolution , There are usually multiple levels of rectangular cells , These units form a hierarchical structure ： Each high-level unit is divided into several lower level units . Statistics about the attributes of each grid cell （ For example, the average 、 Maximum and minimum ） Pre calculated and stored . These statistics are used to answer queries .
• Density based clustering method , Such as DBSCAN
  DBSCAN By checking the ε- Neighborhood to find clustering . If a point p Of ε- Neighborhood contains more than MinPts A little bit , Create one to p New clusters as core objects . then ,DBSCAN Repeatedly look for objects with direct density from these core objects , This process may involve the merging of some density reachable clusters . When no new points can be added to any cluster , End of the process .
• Clustering method based on neural network , Such as SOM
  Self organizing feature mapping network SOFM Also known as self-organizing mapping network SOM, It is a self-organizing competitive neural network , When a neural network accepts external input patterns , Will be divided into different corresponding areas , Each region has different response characteristics to the input mode , And this process is done automatically . Its characteristics are similar to the self-organization characteristics of human brain .
• Graph based clustering method , Such as Normalized cut