An implementation of Tak and a computer opponent in C https://git.l-3.space/ctak/atom/include/nn.h?h=hybrid_rootstrap 2026-09-16T20:02:53Z 2026-09-16T20:02:53Z tslil tslil@posteo.de 2026-09-16T19:10:33Z urn:sha1:1898add0142b70b3fef5af78a6f3d7399a3f2cd5 A little subtle, but in non-self-play games sharing tt entries is incorrect. We must therefore key additionally on the evaulation function itself. 2026-09-10T20:04:40Z tslil tslil@posteo.de 2026-09-10T13:12:41Z urn:sha1:220a4b79e4b6d742a9ca7c278a1962b870500fbc The idea is that the current model is M_theta(board, ply) = "score for player 0", but we can instead create a different model M'_theta(board) = "score for player 0 assuming that it's player 0's move" and for a certain pair of functions sigma, normalise, recover M_theta(board, ply) = sigma(ply) M'_theta(normalise(board, ply)). This pair of functions is sigma(ply) = +/- 1 (parity), and normalise being a conditional interchange of pieces on the board depending on the parity of the ply. The only things stopping this from being completely straightforward is that nn_train.py evaluates M'_theta itself so we need to pass sigma(ply) to the training loop. 6x6 is still untrained. 2026-09-09T20:45:17Z tslil tslil@posteo.de 2026-09-08T10:55:28Z urn:sha1:1c9e266ae25f7f6067806d139e5971ce8513d2f6 2026-09-04T20:03:59Z tslil tslil@posteo.de 2026-09-03T19:17:22Z urn:sha1:1ca3309dad9d412870b2b793f33af7a7ccabd514 TD-lambda and TD-leaf are separated by the observation that minimax with a parameterised leaf heuristic is differentiable* in the parameters. "All" that had to be done was to extract the "principal variation": which one position survived all of the pruning to become the reported value. This was a bit fiddly and uncovered a subtle bug (again) in negamax.c: we were not being complete in our overriding of the TT entry and by omitting key it was possible that later searches would find incorrect values. The more interesting question to me was: how do we introduce enough variety while not corrupting the PV chain? I settled on having selfplay have one of the players be epsilon-random but the other greedy (minimax) and simply report only the greedy portion PV chain. This appears to work well enough in practice and doesn't introduce nonsense training data for the net (we don't force its actions to be nonsensical). I also deciced, perhaps in a departure from tradition, that we had good supervised labels for PVs that happened to be winning states, so the training overrides the targets in that case. Finally rolled back some of the ambition in nn_train.py default arguments. *almost always 2026-09-02T11:13:01Z tslil tslil@posteo.de 2026-08-30T18:46:07Z urn:sha1:241156da6f35cb888253e6b8814e7b61a76b4fc7 NOTE: 6x6 weights are as-yet untrained On the performance side: * switch from linked-list chaining hash table to direct mapping * cache check_win by hash * persist transposition tables across plies (this was a big deal and difficult to puzzle out) * add align directives to the network floats * prepare_nn_input gained a small LUT to avoid branching (i suppose you might call this 1D embedding) * apply mtune=native for the native builds On the correctness side: * Zobrist has now includes the ply parity * the network now predicts something in [-1, 1] using tanh (approximated by polynomials) * the training object is now MSE because we're learning a value estimate * use .clangd instead of compile_commands.json On the training side: * completely forgo human-derived data set * implement td(lambda), self-play, tournaments 2023-01-21T21:00:57Z tslil clingman tslil@posteo.de 2023-01-21T18:30:45Z urn:sha1:72c5056c5aa614d9a86f8b760955cfcf96c44be4 Same network architecture, same training principle. Predictably this is too slow. Also statically allocate state in driver programmes.