Unsupervised Learning: allow us to close the issue when the results can not be foreseen, "unsupervised learning", the tag information of training samples is unknown, the goal is the inherent nature and rules to reveal data by learning unmarked training samples provide a basis for further data analysis.
9.1 Cluster mission
Common unsupervised learning tasks: clustering, density estimation, anomaly detection.
Clustering trying to sample data set is usually divided into several disjoint subsets , each subset is called a "cluster."
Clustering algorithms: clustering algorithm it or ethnic group is divided into a series of data points in the group, so that the same cluster of data points closer to the point than the other clusters. Spacer groups having similar traits, assigned to the cluster.
Clustering both as a separate process for finding the internal distribution of structured data, but also as a precursor process of classification and other learning tasks.
9.2 Performance Metrics
Cluster performance measure known as "effectiveness indicators", performance metrics and supervised learning effect is similar.
What kind of clustering results squatters: "Like attracts like," we hope "cluster similarity" clustering of high and "inter-cluster similarity" low.
Clustering performance measure roughly two types: one is