[MATH-1548] Inefficient computation of SOM quality standard indicators

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Done
• Affects Version/s: 3.6.1
• Fix Version/s: 4.0
• Assignee: Gilles Sadowski
• Reporter: Gilles Sadowski
• Created: 26/Jun/20 01:24
• Updated: 28/Jun/20 23:01
• Resolved: 28/Jun/20 23:01

Description

Package o.a.c.m.ml.neuralnet.twod.util contain classes for defining various measures of the quality of a self-organizing feature map (SOM).
The current design can be quite inefficient when the number of data and/or the number of neurons are large, since the same procedures (e.g. finding the neuron whose features most closely matches some data) are repeated for each quality indicator.
It is proposed to compute all "standard" indicators at once (in class NeuronSquareMesh2D).