PVS & Embla

[flok]

Hi,

I tried that but:

Code: Select all

Rank Name       Elo    +    - games score oppo. draws
   1 dorpsgek   134    5    4 28434   70%   -44   16%
   2 trunk       87    5    6  9524   44%   134   20%
   3 pvs2b2      34    6    6  9474   37%   134   16%
   4 lmrpvs    -254    7    8  9436    8%   134   13%

As you see the to pvs variants I tried still play far worse than the non-pvs version (trunk).

[abulmo2]

I guess there is one or several bugs in other places.
Within the loops (in both search & quiescence), the initialization to 0 of the score variable looks suspect to me. In some (rare ?) cases, bestval will be set to 0 instead of a more legible negative score.

Code: Select all

			int score = 0;
#ifdef USE_TT
			if (!addToHistory(sd -> history, newHash.newHash))
#endif
			{

should be replaced by

Code: Select all

			int score = -infinite; // anything <= bestval
#ifdef USE_TT
			if (!addToHistory(sd -> history, newHash.newHash))
#endif
			{

[flok]

The score will be 0 if a position repeats. This is to prevent 3-repeats.

[abulmo2]

Sorry, your code is obviously too clever for me

[AlvaroBegue]

Sorry, your code is obviously too clever for me

That is a problem. Clever code hides bugs. Code should be as clear as possible.

Of course I am guilty of writing clever code myself, but I do make some effort to keep things simple.

The way I handle repetitions is to call a function that checks for repetition and 50-move rule before the recursive call to negamax:

Code: Select all

      score = board.draw() ? 0 : -negamax(-beta, -alpha, depth - 1);

[abulmo2]

That is a problem. Clever code hides bugs. Code should be as clear as possible.

Of course I am guilty of writing clever code myself, but I do make some effort to keep things simple.

The way I handle repetitions is to call a function that checks for repetition and 50-move rule before the recursive call to negamax:

Code: Select all

      score = board.draw() ? 0 : -negamax(-beta, -alpha, depth - 1);

My programs detect draws (3-fold repetition, 50-moves rule and lack of material) at the start of the negamax function, ie after the recursive call.

Code: Select all

	if (board.isDraw) return 0;

Before or after a function call is not important, but naming such a function with the word "draw" inside is.

I am impressed by the work that is behind Embla, that contains many more things than in Amoeba, like syzygy table support, mysql/sqlite3 opening books, multithreading, etc. But I am also disappointed by the weakness of the engine, that is much weaker than my bare engine https://github.com/abulmo/Dumb. I am convinced that some bugs or poor implementation are hidden in the code and prevent it to progress.

[flok]

Possibly yes, but I have no idea where.
Note that the movegen is a bit slow, that doesn't help either.

[Henk]

Posting on this website makes your code also worse is my experience. Best is to stick to your own bad coding ideas.

[Sven]

Posting on this website makes your code also worse is my experience. Best is to stick to your own bad coding ideas.

So maybe not posting for six months would give +600 Elo for Skipper?

[Henk]

Posting on this website makes your code also worse is my experience. Best is to stick to your own bad coding ideas.

So maybe not posting for six months would give +600 Elo for Skipper?

It would already be an improvement if Skippers rating stayed equal.