Playground on various subjects.

I can't play sudoku, but can I write program that can play sudoku?

Current solvers are:

• brute solver, which probably won't find solution till sun shines
• naive filtering (check is state is consistent before processing), which should be better
• forward filtering - remove unconsistent values in neighour nodes - blazingly fast (simple problem in 300 tries, hard in tens of thousands)
• arc consistency checking

Further notes in subdirectory sudoku

Playing with MDP. AI learning to play classical PONG game. Available in browser Pong AI

Further notes in subdirectory mdp

Attempts to model heart rythm with HMM for inference on pulse state.

Currently exercises with HMM algorithm in Crooked Casino from Bioinformatics Algorithms course.

AI learning to play Google TRex. Attempts to learn to play using 2 layer backpropagated network. In a kind of a reinfored learning (simplified Pong from Pixels)

Further notes in subdirectory neural

Exercises with Kohonen Self Organizing Maps. Mapping colour space to 2D graph, maping linear functions, etc. More in subdirectory kohonen

Two dimensional kohonen map learning color space. CLICK TO PLAY

To be done:

• space learning "robotic arm"
• balancing robot

Graphics renderer capable of drawing geometry primitives: dots, lines, boxes, curves. First implementation is to be 2D.

Visualization tool needed for other projects.

Bayesian Network computer for finding genetic traits in ancestry tree (e.g. child blood type).

Plays with one of the simples neural models.