Didactical neural nets

[chesskobra]

I would like to learn how neural nets have replaced the hand tuned values of the engine's evaluation functions. However, present neural nets have grown too big and I am afraid also that the complexity of generating them would be too much for me (and for my hardware). So, I thought that I could start from a simpler position, say rook + king vs. king alone. Is there already someone who has done something similar for didactical purposes? Also, if I should build it myself where I should start from?

I think it should be possible to train simple endgames. This free book https://github.com/asdfjkl/neural_network_chess shows how to train an NN for a simple game such as hexapawn. Something like that you could do for KR vs K. But even more interesting would be to use a tablebases of up to 4 or 5 piece endgames for training an NN. Has anybody tried training NNs on tablebases? It would of course not play all endgames in the TB perfectly, but if it played a large fraction of them (not even necessarily finding the shortest win or draw) that would be cool.