ART diagrams
============

This page provides process flow-charts for generalized ART algorithms. These algorithms
are not specific to any particular variant (e.g., Fuzzy, Gaussian, etc.) and provide a
high-level overview of how the algorithms work.

This page is not yet comprehensive of all algorithms but will continue to grow.

.. contents::
   :local:
   :depth: 1

ART Process Flow
----------------

.. figure:: ../diagrams/basic-art-diagram.svg
   :name: basic-art-diagram
   :alt: ART process flowchart
   :align: center
   :width: 100%

   Generalized ART process flow from raw data to weight updates.

SimpleARTMAP Process Flow
-------------------------

.. figure:: ../diagrams/simple-artmap-diagram.svg
   :name: simple-artmap-diagram
   :alt: Simple ARTMAP process flowchart
   :align: center
   :width: 100%

   Generalized Simple ARTMAP process flow from raw data to weight updates. SimpleARTMAP
primarily functions by introducing a second "check" to the vigilance phase of the
standard ART algorithm. This second vigilance enforces a 1-to-many mapping from cluster
to class label and thus learns a functional mapping. If this second check fails, the
first vigilance check is adapted according to the match-tracking function
(which may either relax or restrict the vigilance check).

ARTMAP Process Flow
-------------------

.. figure:: ../diagrams/standard-artmap-diagram.svg
   :name: simple-artmap-diagram
   :alt: ARTMAP process flowchart
   :align: center
   :width: 100%

   A generalized, standard ARTMAP process flow from raw data to weight updates.
Standard ARTMAP differs from SimpleARTMAP in that it allows the y-side data to be
real-valued vectors rather than just labels. ARTMAP effectively functions by allowing
the y-side data to be clustered in an unsupervised fashion and then feeding the
resulting cluster labels along with the x-side data to SimpleARTMAP.