Having Fun with Self-Organizing Maps

Self-Organizing Maps (SOM), or Kohonen Networks ([1]), is an unsupervised learning method that can be applied to a wide range of problems such as: data visualization, dimensionality reduction or clustering. It was introduced in the 80' by computer scientist Teuvo Kohonen as a type of neural network ([Kohonen 82],[Kohonen 90]). In this post we are going to present the basics of the SOM model and build a minimal python implementation based on numpy. There is a huge litterature on SOMs (see [2]), theoretical and applied, this post only aims at having fun with this model over a tiny implementation. The approach is very much inspired by this post ([3]).