TalkChess.com

Looks like a draw with many 7 piece TBs also. 0.00.

Jhoravi wrote: ↑

Sat Jan 30, 2021 8:57 am

That is possible even today as long as the engines favor complexity over best moves.

Maybe somebody could train a net for knight odds against a weaker opponent. It would be interesting to see how it would try to complicate positions.

Are there are any games to check out? It would be fun to go through them to see what happened. Of course, it shouldn't be a purposefully bad opening.

I don't know what would work well in your case. I did an experiment with a specific ending to check how well a custom-trained net can predict whether a position is a mate or not and the net could get 98% right vs 90% for SF NNUE. Based on this, I am sure that it's possible to improve SF NNUE by some...

Because of its futile. It's depend on the inputs. NN learn this. Please read more about neural networks. Smug and wrong, not a good combo. It might easily turn out that a different board representation as the input to the NN or a different NN architecture or a different number of parameters is supe...

Seems pretty similar to my suggestion from February: http://talkchess.com/forum3/viewtopic.php?f=2&t=72988. I'm for it.

There is nothing great. https://www.reddit.com/r/cpp/ is ok if you want to talk about the standard.

Thanks for the posts, good info. I will do some reading and playing around with code and see where it goes.

How are you going to "mark" a position "as having the true value stored" without encoding this in the value you store for the position? One way could be like I wrote earlier: true value (1 bit, ignoring cursed wins/losses), gap (maybe 7 bits, maybe depends on the endgame), true value, gap... A sort...

Thanks for the post. Since the heuristic won't give distance to mate, you can only do w/d/l with this, so the table would have 6 possible values w/d/l/w50/l50/<search> with <search> being a large majority of the results. I think that cursed win/blessed losses are rare enough that it would be sane to...