This Repository Explains various Unsupervised Learning Techniques and their Disadvantages

Uses of Unsupervised Learning

• Density Estimation
• Finding Latent Variables
• Dimensionality Reduction

K Means Clustering

Algorithm:

Initialize: Pick K Random Points to be the Cluster Centers.

While not converged:
Assign each points to the nearest cluster center.
Recalculate each cluster center from points that belong to it.

Soft K Means Clustering

Algorithm

Visualization of Convergence of Soft Kmeans

K_Means Fails on These Distribution

Donut

Multivariate Gaussian

Gaussian with different Density

Disadvantages of K Means

1. You have to choose K
2. Stuck in Local Minima
3. Sensitive to Initial Configuration
4. Can't Solve the Donut Problem
5. Doesn't Take into account Density of the cluster

Hierarchial Clustering

Ward Linkage

Complete Linkage

Single Linkage

Gaussian Mixture Model

Algorithm

1. Calculate Responsibilities

2. Calculate Model Parameters

Convergence in Gaussian Mixture Model