An implementation of Tak and a computer opponent in C https://git.l-3.space/ctak/atom/include/negamax.c?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. 2026-09-12T14:47:15Z tslil tslil@posteo.de 2026-09-11T17:36:10Z urn:sha1:a361728f39812e25599323c52510c0eb3131bd90 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 2026-05-05T21:16:39Z tslil tslil@posteo.de 2026-05-05T21:06:02Z urn:sha1:696c3808220ea9d8c604939894a2d509f05cc4c6 we needed to be saving the alpha before the search loop in which we modified it. At this point, though i can't remember, i expect that this is the cause of whatever "instability" the comment was talking about that would have prevented us from using >= in the TT lookup. 2023-01-29T20:52:40Z tslil clingman tslil@posteo.de 2023-01-29T19:49:04Z urn:sha1:2e5ffeaf928626c81ea3147b57b7ee2c660ebfae - attempts to keep the same move ordering as the list method - saves ~170msec on a depth 7 search for a given board configuration - there is room to improve the pre-allocation size estimates, these bounds are not obviously tight and it may or may not be faster to have tighter bounds or even some form of estimation 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. 2023-01-19T21:40:24Z tslil clingman tslil@posteo.de 2023-01-19T21:40:24Z urn:sha1:24d590fb7fe501a59c9ee0268a956a6fbf4fa092 Given the approximate nature of the evaluations and truncated tree searches, the result of negamax will always be sensitive to the particulars of the depth bounding and the conditions for referring to precomputed values. The tradeoff here is a slight performance penalty (~200ms at depth 6 on my old Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz), but a greatly increased opponent strength as compared to `>=`, and the old convolutional network (flawed as the implementation was). 2023-01-16T11:25:47Z tslil clingman tslil@posteo.de 2023-01-15T20:31:00Z urn:sha1:ef785da721a47e6d13aa854c795d2a9e269285b0 Previously the code base assumed that there was a single, global game state which was the implicit target of all actions taken. Looking ahead at architectural improvements, this has now been (almost entirely) made explicit and functions take tak_state_p where necessary (and also where unnecessary). Two important fixes to actions.c were made: - Previously when generating the possible stack moves, stack height overflows (> 15) were not taken into account and this resulted in the tree search corrupting the board state. Now action search does not list all legal actions, rather the subset of these encodeable by the implementation. - The check for crushing on a stack move was incorrect (too strict), and this resulted in many legitimate moves being igonored. Finally, in other changes, weights have also been improved by training all games instead of some subset for chosen players, and clang-format was run on the codebase.