Computer Chess Club
https://talkchess.com/
That is very surprising.Albert Silver wrote: ↑Mon Jul 23, 2018 5:39 pm I should add that one can use Leela weights too. I have not tested beyond seeing they work.
Why would it be surprising? LC0 search is anyway total crap as it A0 search, and ppl currently involved in LC0 development clearly have not much clue how to improve or even only evaluate search performance.Robert Pope wrote: ↑Tue Jul 24, 2018 4:02 amThat is very surprising.Albert Silver wrote: ↑Mon Jul 23, 2018 5:39 pm I should add that one can use Leela weights too. I have not tested beyond seeing they work.
Could you give it a try, Albert?Albert Silver wrote: ↑Mon Jul 23, 2018 5:39 pm I should add that one can use Leela weights too. I have not tested beyond seeing they work.
Because the weights are tied to the specific neural network implementation format of LC0. In order for them to function in another engine, that engine has to have the exact same input planes, with the same orientation, in the exact same order, in the exact same format. Otherwise, it would be like porting Stockfish evaluation weights into another engine, except you are assigning them at random. Not even all LC0 neural nets are compatible across versions of LC Zero.Milos wrote: ↑Tue Jul 24, 2018 4:59 amWhy would it be surprising? LC0 search is anyway total crap as it A0 search, and ppl currently involved in LC0 development clearly have not much clue how to improve or even only evaluate search performance.Robert Pope wrote: ↑Tue Jul 24, 2018 4:02 amThat is very surprising.Albert Silver wrote: ↑Mon Jul 23, 2018 5:39 pm I should add that one can use Leela weights too. I have not tested beyond seeing they work.
I highly doubt this too. For me the Scorpio NN files are completely different.Robert Pope wrote: ↑Tue Jul 24, 2018 4:02 amThat is very surprising.Albert Silver wrote: ↑Mon Jul 23, 2018 5:39 pm I should add that one can use Leela weights too. I have not tested beyond seeing they work.
I don't think you understand how MCTS works. In Scorpio it is using just value network as direct rollout replacement (i.e. as eval). Policy network is not used to guide the search. MCTS is completely agnostic to the way evaluation only is performed (A/B rollout, handcrafted NN eval, LC0/A0 NN eval, etc.). Format, number of layers, input planes of NN and all the internals are irrelevant for search.Robert Pope wrote: ↑Tue Jul 24, 2018 5:36 pmBecause the weights are tied to the specific neural network implementation format of LC0. In order for them to function in another engine, that engine has to have the exact same input planes, with the same orientation, in the exact same order, in the exact same format. Otherwise, it would be like porting Stockfish evaluation weights into another engine, except you are assigning them at random. Not even all LC0 neural nets are compatible across versions of LC Zero.Milos wrote: ↑Tue Jul 24, 2018 4:59 amWhy would it be surprising? LC0 search is anyway total crap as it A0 search, and ppl currently involved in LC0 development clearly have not much clue how to improve or even only evaluate search performance.Robert Pope wrote: ↑Tue Jul 24, 2018 4:02 amThat is very surprising.Albert Silver wrote: ↑Mon Jul 23, 2018 5:39 pm I should add that one can use Leela weights too. I have not tested beyond seeing they work.
So for them to actually work in another engine is either a mind-boggling coincidence, or intentional copying of the format (not that there's anything wrong with that [/seinfeld]). IIRC, ScorpioMCTS came first, so it would be surprising to find that its neural net layout is exactly the same.
Except....the value network is an output of the neural network, and is simply a series of transformations of the inputs and the weights. So of course the format of the neural network matters! If you don't have the same format, then the value network that gets output is junk. How MCTS uses that to guide the search is what is irrelevant, since it's garbage information.Milos wrote: ↑Tue Jul 24, 2018 7:12 pm I don't think you understand how MCTS works. In Scorpio it is using just value network as direct rollout replacement (i.e. as eval). Policy network is not used to guide the search. MCTS is completely agnostic to the way evaluation only is performed (A/B rollout, handcrafted NN eval, LC0/A0 NN eval, etc.). Format, number of layers, input planes of NN and all the internals are irrelevant for search.