The first neural layer. Test3 takes the camera's pixels and folds them — through a reversible, nested self-organizing map — into a structured intermediate representation.
A Java/Swing visual lab. Feature and image vectorizers turn input into vectors; a BaseSOM → MetaSOM hierarchy organizes them; ReversibleSOMHierarchy keeps the mapping invertible (forward: vector → winning node; backward: node → vector), so the layer can be read in both directions.
Several Swing panels — IndividualSOMPanel, SOMVisualizationPanel, SOMHierarchyVisualizer — render the maps live as they train, so you can watch the high-dimensional input settle into a 2-D organization. It produces the intermediate vectors the second layer consumes.
Nf3the knight is the only piece that leaps the board — this is the only layer that folds the plane onto itself.
O-O ↔ O-O-Oevery forward map remembers the door it came through, so it can always castle back.
…the pawn structure is the positionorganize the small things and the rest of the game declares itself.