1. Oliver Lomp
  2. SOMPlugin

Overview

HTTPS SSH

README

This plugin provides steps related to self-organizing maps and other learning methods for cedar.

How do I get set up?

To get started with this plugin, copy the cedar.conf.example to cedar.conf and enter the path to the cedar version you want to use, either in CEDAR_HOME if you use a local version of cedar, or in CEDAR_HOME_INSTALLED if you are using a version of cedar installed with make install.

Then cd into the plugin folder and type

mkdir build
cd build
cmake ..
make

The plugin is then found in the build directory and can be loaded into cedar as usual.

What does the plugin provide?

Steps:

  • KohonenMap: this is continuous-time version of the Kohonen SOM. The equations are largely the same as in the original SOM, but learning no longer happens iteratively.
  • HebbianLearning: This implements standard hebbian learning.
  • MatrixToVector and VectorToMatrix: Canonically converts two-dimensional matrices to one-dimensional ones and back again.
  • WinnerTakeAll: Sets all but the maximum entry in a matrix to zero, and the maximum to one.

Data & Plots:

  • CenterData and Plot: the KohonenMap uses this data to represent its clusters. This type of data can be plotted by the CenterDataPlot added in the plugin.

Known Issues

  • The center data plot sometimes does not notice when the center data in an SOM changes, e.g., when learning is reset or when an architecture is loaded. Reopening the plot will fix this.