nps of Crafty same as Robbo/SF

[bob]

But a simple question. You take a program that does that, and inside the search loop you add new code to make and unmake every last move, do an evaluation, etc. And out of that set of moves, you decide that only one deserves searching to the next ply. Did you _really_ think that program's nps just dropped by some big factor, because you added forward-pruning? There's been a years-old argument about hash hits, since with a hash hit you effectively re-search that entire sub-tree instantly. And one could make a legitimate argument that hashing is an optimization that saves time but searches the same tree.

The only reason most keep an nps is to see the effect and effectiveness when changes are made to a program. I earlier found my nps too low as compared to those of Robbolitto and stockfish; so now I did some optimization and it seems I could make further changes that could make my program faster. One thing my low nps hint at is that I am not doing my evaluation correctly; I now have an idea of how better to re-write my evaluator. So this is the practical use of keeping an nps. In my case, comparison with the nps of other programs is helpful.

Rasjid

BTW, in looking back, someone noticed a big drop in NPS (in Crafty) when I added the new forward pruning stuff (not sure when this happened, but several versions back.) Some discussions here led me to make the change. And the good thing was, the version prior to forward-pruning, and the version after the change in the way nodes are counted, but produced nearly identical NPS numbers again, which suggests that this is a reasonable way to count, even if it may not be compatible with everyone.

[mcostalba]

even if it may not be compatible with everyone.

I am not a native speaker so perhaps I am wrong, but I would think should be written. "even if is not compatible with anyone"

Just a joke ! Of corse every one is free to write what he prefers on his nps, even to take the nps and multiply x8 before to printf them...hmmmm...now that I think better...hmmm...why not ?

[bob]

even if it may not be compatible with everyone.

I am not a native speaker so perhaps I am wrong, but I would think should be written. "even if is not compatible with anyone"

Just a joke ! Of corse every one is free to write what he prefers on his nps, even to take the nps and multiply x8 before to printf them...hmmmm...now that I think better...hmmm...why not ?

I don't think my node count is significantly different from anyone until one gets to the issue of make/prune/unmake or make/search/unmake.

However, in the classic AI sense, node == position, and certainly if one makes a move and does any analysis on the resulting position, that position would have to be considered a new node. Because that is all that there is. Nodes, and moves, where two nodes are connected by an arc that represents the move necessary to get from N to N'...