TalkChess.com

I would do something like this: #include <iostream> typedef unsigned long long u64; void print_bb(u64 x) { for (int row = 7; row >= 0; row--) { for (int col = 7; col >= 0; col--) std::cout << ((x >> (row*8+col)) & 1); std::cout << '\n'; } } inline u64 N(u64 x) { return x << 8; } inline u64 W(u64 x) ...

1-0 for white. I use the same format of the "Result" tag in PGN.

Offhand, I'd say maybe, but that is a very speculative maybe. One cannot remove tactics from the equation, so oversights in its calculations will affect its decisions. An argument such as ' 'this would be a great move if.... it didn't lose a piece" holds no water in my book. Once Capablanca was ask...

Albert Silver wrote: ↑

Tue May 22, 2018 4:00 pm

I am having trouble with the UCI options though when configuring the engines. Some are more than one word and have spaces. Any idea how best to solve this?

Why is that a problem? The UCI protocol allows for multiple words in the name of options.

Do you really need more than 1 byte for depth? In RuyDos the maximum depth is 64, and even if I had fractional values, I think I wouldn't use more than 8 bits. There are only three possible values for the flag, so 8 bits is very wasteful. Also 8 bits for age is a lot. You can combine those two field...

For policy training you also need move probabilities. Relative frequencies for all legal moves per position. And who can tell that these values are right. No, for policy training you just need moves. Take any collection of games (I believe they don't need to be of particularly high quality or evenl...

I can give you two related things. The first one is the file I used to tune RuyDos a couple of years ago: https://bitbucket.org/alonamaloh/ruy_tune/raw/74dd679130c5c9d3c186729fcba0dc2638dcef4d/data/quiescent_positions_with_results It looks like this: 3r4/4k3/8/5p1R/8/1b2PB2/1P6/4K3 b - - 1-0 3nk2r/r...

You could also check if the hash move is legal in the current position, as further verification. One final twist to that trick consist of storing the move XORed with the additional bits of the hash key that you are not saving or using for addressing. That way the probability that the move you read b...

I keep actual scores in centipawns but that is irrelvant to the minmax/averaging stuff. Sorry, I haven't followed the whole thread, but this statement makes no sense to me. Since the conversion from "expected result" space (i.e., [-1,1]) to "score in centipawns" is not linear, you can't expect aver...

Daniel Shawul wrote:[...] I have just finished writing the c++ api to integrate tensorflow so time for expermentation on search with NN eval ..

How did you do that? Did you use something like tensorflow_cc (https://github.com/FloopCZ/tensorflow_cc)?