Self Organising Map learning parameters

Theta/Momentum - This affects the initial distance a neuron is moved towards a data point when the map is adapted to fit the data set in the training phase of the SOM. The Momentum box gives an opportunity to set the "friction"-rate when moving a neuron. The Momentum is constant during the training phase.

Phi/Momentum - This sets the amount of "pull" working between neurons, in other words how much the neurons should affect each other. The Momentum box sets the "stiffness" of the links between neurons. The Momentum is constant during the training phase.

In the beginning when theta is large the neurons are moved big steps (theta) and affect its neighbour neurons a lot (phi) to learn "global" patterns. After a while it becomes more important for the neurons to adjust to local environments. The momentum then reduces theta and phi. When theta and phi are both 0, the learning period is over.

Sometimes it is better to keep the elasticity during local adjustments and therefore keep phi a bit longer (larger momentum on this one). If momentum is over 1.0, theta and phi will increase for each iteration, which makes no sense. Theta larger than 1.0 will move the neuron past the data point pulling on it, which makes no sense either. All these values must therefore be between 0.0 and 1.0