Right now my engine can reach up to ply 8 (not including Q search) in about 20 seconds on avg.
The only saving features I have is:
MVV/LVA move ordering and killer.
My question is does 8 ply at 20 seconds sound right? Is it too slow?
What type of savings can I expect with each?
Transposition Tables
PV-Search Pruning
Null Move
Aspiration windowing
futility
What do you guys recommend I should do next now?
Thanks in advance!
Want to know what savings to expect with pruning methods...
Moderators: hgm, Rebel, chrisw
-
- Posts: 154
- Joined: Tue May 17, 2011 8:12 pm
-
- Posts: 892
- Joined: Sun Nov 19, 2006 9:16 pm
- Location: Russia
Re: Want to know what savings to expect with pruning methods
1) Qsearch
2) Iterative deepening with PV-move first
3) Transposition Table
4) Null-move
2) Iterative deepening with PV-move first
3) Transposition Table
4) Null-move
-
- Posts: 1822
- Joined: Thu Mar 09, 2006 11:54 pm
- Location: The Netherlands
Re: Want to know what savings to expect with pruning methods
How many positions per second is it searching? Or better: makemoves per second i'd argue.voyagerOne wrote:Right now my engine can reach up to ply 8 (not including Q search) in about 20 seconds on avg.
The only saving features I have is:
MVV/LVA move ordering and killer.
My question is does 8 ply at 20 seconds sound right? Is it too slow?
What type of savings can I expect with each?
Transposition Tables
PV-Search Pruning
Null Move
Aspiration windowing
futility
What do you guys recommend I should do next now?
Thanks in advance!
-
- Posts: 154
- Joined: Tue May 17, 2011 8:12 pm
Re: Want to know what savings to expect with pruning methods
I already have 1 and 2.Aleks Peshkov wrote:1) Qsearch
2) Iterative deepening with PV-move first
3) Transposition Table
4) Null-move
So I assume I should do 3 next...
-
- Posts: 154
- Joined: Tue May 17, 2011 8:12 pm
Re: Want to know what savings to expect with pruning methods
Its averaged around 2500kns. It counts only legal moves that are played on the virtual board.How many positions per second is it searching? Or better: makemoves per second i'd argue.
-
- Posts: 1357
- Joined: Wed Mar 08, 2006 10:15 pm
- Location: San Francisco, California
Re: Want to know what savings to expect with pruning methods
Do I read this right combined with other posts in the thread? You're doing 2.5 million nodes per second WITHOUT a transposition table and null move?voyagerOne wrote:Its averaged around 2500kns. It counts only legal moves that are played on the virtual board.How many positions per second is it searching? Or better: makemoves per second i'd argue.
jm
-
- Posts: 154
- Joined: Tue May 17, 2011 8:12 pm
Re: Want to know what savings to expect with pruning methods
Yes that is correct.
Do I read this right combined with other posts in the thread? You're doing 2.5 million nodes per second WITHOUT a transposition table and null move?
jm
I wrote the program in JAVA...and my methods are somewhat unorthodox. Such as each piece has its own bitboard, among other things...
-
- Posts: 1357
- Joined: Wed Mar 08, 2006 10:15 pm
- Location: San Francisco, California
Re: Want to know what savings to expect with pruning methods
Ok, one more question. Are you talking about the speed of your movegen plus make/unmake only (i.e. perft speed), or is this the speed of your full alpha-beta search?voyagerOne wrote:Yes that is correct.
Do I read this right combined with other posts in the thread? You're doing 2.5 million nodes per second WITHOUT a transposition table and null move?
jm
I wrote the program in JAVA...and my methods are somewhat unorthodox. Such as each piece has its own bitboard, among other things...
jm
-
- Posts: 154
- Joined: Tue May 17, 2011 8:12 pm
Re: Want to know what savings to expect with pruning methods
Ok, one more question. Are you talking about the speed of your movegen plus make/unmake only (i.e. perft speed), or is this the speed of your full alpha-beta search?
alpha-beta search speed.
So its something like this...
Generate all moves;
if move is illegal Then
get next move
makeMove;
plyCount++;
-
- Posts: 454
- Joined: Sat Apr 04, 2009 6:44 pm
- Location: Bulgaria
Re: Want to know what savings to expect with pruning methods
I don't see anything unusual. Right now, my program with magics, simple /almost empty/ evaluation and basic search does 4000 kn/s /midgame/.JVMerlino wrote:Do I read this right combined with other posts in the thread? You're doing 2.5 million nodes per second WITHOUT a transposition table and null move?voyagerOne wrote:Its averaged around 2500kns. It counts only legal moves that are played on the virtual board.How many positions per second is it searching? Or better: makemoves per second i'd argue.
jm
What settles that down is when the node get filled with some useful computation.
p.s. Don't forget the exact architecture!