Movei Engine: Need Help Loading Specific Hash

Questions and comments related to CCRL testing study

Movei Engine: Need Help Loading Specific Hash

Postby Chuck Wilson » Mon Oct 16, 2006 11:59 am

Hello! I know Uri mentioned that he is not so concerned with hash table settings but I would like to test his engine evenly against others. I'm currently trying to run Movei 0.08.295 but it only loads 1 MB hash tables.

I tried editing the Movei changes file, which only has one setting for loghash, but changing that hasn't helped. Does anyone know what the setting must be to get the engine to load x MB of hash? :? Thanks!

Cheers,

Chuck
Chuck Wilson
 
Posts: 540
Joined: Fri Oct 13, 2006 2:53 pm
Location: Annapolis, Maryland, USA

Re: Movei Engine: Need Help Loading Specific Hash

Postby Uri Blass » Mon Oct 16, 2006 2:08 pm

Chuck Wilson wrote:Hello! I know Uri mentioned that he is not so concerned with hash table settings but I would like to test his engine evenly against others. I'm currently trying to run Movei 0.08.295 but it only loads 1 MB hash tables.

I tried editing the Movei changes file, which only has one setting for loghash, but changing that hasn't helped. Does anyone know what the setting must be to get the engine to load x MB of hash? :? Thanks!

Cheers,

Chuck


Movei need special change file to use only 1 mbyte hash because the default without editing the change file means 64 mbytes.

loghash 16 in the change file means 1 mbytes hash
loghash 17 in the change file means 2 mbytes hash
loghash 18 in the change file means 4 mbytes hash
loghash 19 in the change file means 8 mbytes hash
loghash 20 in the change file means 16 mbytes hash
loghash 21 in the change file means 32 mbytes hash
loghash 22 in the change file means 64 mbytes hash
loghash 23 in the change file means 128 mbytes hash

I am not sure if bigger is always better for movei.

Uri
Uri Blass
 
Posts: 40
Joined: Tue Jun 20, 2006 12:44 pm

Postby Chuck Wilson » Mon Oct 16, 2006 10:04 pm

Thanks Uri, I'll try these settings. I'll run the test with 128 MB Hash and we'll see what happens.

Cheers,

Chuck Wilson
Chuck Wilson
 
Posts: 540
Joined: Fri Oct 13, 2006 2:53 pm
Location: Annapolis, Maryland, USA

Postby Ray » Tue Oct 17, 2006 6:38 am

Coincidentally, I was playing around with movei under Arena, and also wasn't getting the right amount of hash. By reading the read-me file I managed to get 128MB by having

loghash 22
loghasheval 22

Is this the same as "loghash 23" or different ?
Ray
 
Posts: 14725
Joined: Sun Dec 18, 2005 6:33 pm
Location: U.K.

Postby Chuck Wilson » Tue Oct 17, 2006 1:33 pm

Uri's advice worked great, so I must thank him for that. I've completed the match between Movei and Amateur and already posted it (it was 30 games at 40/4 - blitz). Movei did win 18.5-11.5. More games are sure to follow.

Cheers,

Chuck
Chuck Wilson
 
Posts: 540
Joined: Fri Oct 13, 2006 2:53 pm
Location: Annapolis, Maryland, USA

Postby Ray » Tue Oct 17, 2006 3:45 pm

Thanks - yes Uri's advice is perfect and simple - but I was trying to find a solution * before * he posted this here

I just want to find out if the settings I found are OK and achieve the same thing, otherwise I have to throw away the games I've run with those settings :)
Ray
 
Posts: 14725
Joined: Sun Dec 18, 2005 6:33 pm
Location: U.K.

Postby Uri Blass » Tue Oct 17, 2006 4:05 pm

Ray Banks wrote:Coincidentally, I was playing around with movei under Arena, and also wasn't getting the right amount of hash. By reading the read-me file I managed to get 128MB by having

loghash 22
loghasheval 22

Is this the same as "loghash 23" or different ?


It is the same size but not the same.

loghash is one type of hash
loghasheval is another type of hash.

loghasheval is used only to save time in calculating the evaluation and bigger is always faster but not by much(increasing loghasheval by 1 may make the program not more than 1% faster).

I am not sure if bigger is always better for loghasheval and the change there is not a simple speed improvement.

I can add that for the FRC rating list it is not important and movei as a UCI engine simply ignores all the information about loghash and use the hash that it gets from shredder gui.

Uri
Last edited by Uri Blass on Tue Oct 17, 2006 4:09 pm, edited 1 time in total.
Uri Blass
 
Posts: 40
Joined: Tue Jun 20, 2006 12:44 pm

Postby Ray » Tue Oct 17, 2006 4:09 pm

Thanks Uri - I'll re-run my games.

They were pretty much just a test anyway, first time I've run games under Arena so I wanted to make sure it would run a batch of games, and it did just fine :)

Arena looked a bit scary at first, but once I found my way round round the menus etc it all made sense
Ray
 
Posts: 14725
Joined: Sun Dec 18, 2005 6:33 pm
Location: U.K.

Postby Uri Blass » Tue Oct 17, 2006 4:13 pm

Ray Banks wrote:Thanks Uri - I'll re-run my games.

They were pretty much just a test anyway, first time I've run games under Arena so I wanted to make sure it would run a batch of games, and it did just fine :)

Arena looked a bit scary at first, but once I found my way round round the menus etc it all made sense


Did you test movei FRC?

I ask because in that case you can test it as a winboard engine or as a UCI engine and time managemant is not exactly the same

Not that as UCI engine the information in the ini file about hash is not important because it get the hash size from the gui.

Uri
Uri Blass
 
Posts: 40
Joined: Tue Jun 20, 2006 12:44 pm

Postby Ray » Tue Oct 17, 2006 4:28 pm

Uri Blass wrote:Did you test movei FRC?

I ask because in that case you can test it as a winboard engine or as a UCI engine and time managemant is not exactly the same

Not that as UCI engine the information in the ini file about hash is not important because it get the hash size from the gui.

Uri


I tested movei FRC under Arena, yes, both as winboard and UCI.
As a UCI engine, it still got the wrong hash - I had 128MB set as a common setting for UCI engines but it didn't seem to pick it up. I'll try again tonight. (it was OK under Shredder GUI)

I'm looking at running FRC matches under Arena now. I would run movei as UCI for consistency. From my existing list, Pharaon and Hermann also should work under Arena. I can run these against Baron, Naum and maybe others which are Arena-specific
Ray
 
Posts: 14725
Joined: Sun Dec 18, 2005 6:33 pm
Location: U.K.

Postby Ray » Tue Oct 17, 2006 4:44 pm

I have to say, Arena is amazing in that it can run winboard engines natively as well as UCI
Ray
 
Posts: 14725
Joined: Sun Dec 18, 2005 6:33 pm
Location: U.K.

Postby Shaun » Tue Oct 17, 2006 4:47 pm

Uri,

I am testing 0.08.373 for you via wb2uci under F9 GUI (because this is how I had tested 0.08.295)

Should I use UCI instead? If you think it may be intersting I can always run both and seperate out the results.

Also

For 0.08.295 I ran with

loghash 23 - for 128mb

under 0.08.373 I am running with

loghash 22 loghasheval 22 - as this came in the supplied ini and gave me 128mb memory usage.

One of the great things about blitz is you can get a statistically sensible number of games in a reasonable amount of time.

Please let me know if there are any particular setting you would like tested - I can always combine latter if they turn out to make no difference to avoid cluttering the rating list.

All the best

Shaun
User avatar
Shaun
 
Posts: 6739
Joined: Sat May 13, 2006 3:24 pm
Location: Brighton. UK

Postby Ray » Tue Oct 17, 2006 6:09 pm

I must have done something wrong before - it is now getting the hash correctly from the Arena GUI for UCI...

I have an FRC match now underway for the ratings list, Movei 366 UCI vs The Baron 1.7.0 :)
Ray
 
Posts: 14725
Joined: Sun Dec 18, 2005 6:33 pm
Location: U.K.

Postby Uri Blass » Tue Oct 17, 2006 11:00 pm

ShaunBrewer wrote:Uri,

I am testing 0.08.373 for you via wb2uci under F9 GUI (because this is how I had tested 0.08.295)

Should I use UCI instead? If you think it may be intersting I can always run both and seperate out the results.

Also

For 0.08.295 I ran with

loghash 23 - for 128mb

under 0.08.373 I am running with

loghash 22 loghasheval 22 - as this came in the supplied ini and gave me 128mb memory usage.

One of the great things about blitz is you can get a statistically sensible number of games in a reasonable amount of time.

Please let me know if there are any particular setting you would like tested - I can always combine latter if they turn out to make no difference to avoid cluttering the rating list.

All the best

Shaun


Movei can be used both as winboard engine and as uci engine.
There may be a difference in the time management so I do not claim that movei has the same playing strength in both cases.

You can use more pruning
default is
futilityprune2 30 70 210 490 890 13
futilityprune 160 480 1040 1840 2880
morelatereduction 10

if you reduce the numbers of futilityprune2 or reduce the numbers of futilityprune or increase the number of morelatereduction then you make more pruning.

You can test
futilityprune2 15 35 105 245 445 7

You can also test
futilityprune 80 240 520 920 1440

You can also test
morelatereduction 20



Uri
Uri Blass
 
Posts: 40
Joined: Tue Jun 20, 2006 12:44 pm

Postby Shaun » Tue Oct 17, 2006 11:38 pm

Uri Blass wrote:You can use more pruning
default is
futilityprune2 30 70 210 490 890 13
futilityprune 160 480 1040 1840 2880
morelatereduction 10

if you reduce the numbers of futilityprune2 or reduce the numbers of futilityprune or increase the number of morelatereduction then you make more pruning.

You can test
futilityprune2 15 35 105 245 445 7

You can also test
futilityprune 80 240 520 920 1440

You can also test
morelatereduction 20

Uri


Hi Uri,

You should have some initial results with 0.08.373 (approx 300 games by this weekend)

I assume you would like the settings applied to this version or would you like to see the results first?

I should be able to run one set of settings a week (may be more - but to be safe lets assume one :wink: ).

Not sure if I said before but for tests like this I use the same computer and the same opponents to reduce the variables.

Shaun
User avatar
Shaun
 
Posts: 6739
Joined: Sat May 13, 2006 3:24 pm
Location: Brighton. UK

Postby Chuck Wilson » Wed Oct 18, 2006 10:06 am

Uri,

This is all good information. Once we get some more games on the list for the latest version I'd like to start testing your beta as well. Can you add to what you've already given us by saying what was the first version with UCI support? I prefer to run engines in UCI mode if available because it just gives more easy access to engine settings and more options with time controls. I do also like the Fritz screen better than Shredder or Arena, but I use Shredder and Arena alot because they have better Winboard support. Of course, if you think you're engine might run better in Winboard mode I'll be happy to test it that way.

Regards,

Chuck Wilson
Chuck Wilson
 
Posts: 540
Joined: Fri Oct 13, 2006 2:53 pm
Location: Annapolis, Maryland, USA

Postby Ray » Wed Oct 18, 2006 6:45 pm

As far as I know, 366 was Uri's first UCI version and the one I'm testing in FRC (of course it still plays winboard too)
Ray
 
Posts: 14725
Joined: Sun Dec 18, 2005 6:33 pm
Location: U.K.

Postby Uri Blass » Thu Oct 19, 2006 8:19 am

1)366 was the first version that was tested with UCI support but the UCI support is not perfect and it may lose on time games with 199 moves or more moves because of a bug.
I also added information in the latest version and it shows more output like the move that the program consoders that is not shown in 366

2)time management may be different in uci mode and winboard mode so I do not claim that the playing strength is the same in both cases.

The problem that I have with uci is that the engine does not know the time control and only get time and moves to the next time control.

There was a discussion about it in the following thread:

http://wbforum.vpittlik.org/viewtopic.php?t=5611

3)I do not think that there are more options with time control for UCI engines.

winboard engines can also play different time control if they get the extended level command from arena

Movei can also play under fritz at time control like 40 moves/2 hours+30 minutes/game but I believe fritz does not give it correct information about the time control so I believe that movei cannot use the information to save time and have significantly more than 30 minutes for the rest of the game in case that it could save time by pondering moves correctly.

This is not very important and it may be more important in extreme time control like 40 moves/2 hours+1 minute/rest of the game.

Uri
Uri Blass
 
Posts: 40
Joined: Tue Jun 20, 2006 12:44 pm

Postby Shaun » Thu Oct 19, 2006 2:28 pm

Movei Gauntlet update:

Code: Select all
CCRL-40in4min-3  2006

Movei 0.08.373   - Arasan 9.2 CCRL            13.5 - 5.5    +12/-4/=3    71.05%
Movei 0.08.373   - E.T.Chess 18.11.05         6.5 - 12.5    +4/-10/=5    34.21%
Movei 0.08.373   - Naum 2.0 (32 bit)          5.0 - 14.0    +2/-11/=6    26.32%
Movei 0.08.373   - Petir 3.99d                9.5 - 9.5    +6/-6/=7    50.00%
Movei 0.08.373   - Pseudo 0.7c                7.0 - 12.0    +5/-10/=4    36.84%
Movei 0.08.373   - Scorpio 1.8.1 (32 bit)     4.0 - 15.0    +2/-13/=4    21.05%
Movei 0.08.373   - Scorpio 1.8.2 (32 bit)     6.5 - 12.5    +5/-11/=3    34.21%
Movei 0.08.373   - The Baron 1.7.0            9.0 - 10.0    +6/-7/=6    47.37%
Movei 0.08.373   - Thinker 4.7a               5.0 - 14.0    +3/-12/=4    26.32%
Movei 0.08.373   - WildCat 6                  12.0 - 7.0    +10/-5/=4    63.16%
User avatar
Shaun
 
Posts: 6739
Joined: Sat May 13, 2006 3:24 pm
Location: Brighton. UK

Postby Uri Blass » Sat Oct 21, 2006 12:49 am

ShaunBrewer wrote:Movei Gauntlet update:

Code: Select all
CCRL-40in4min-3  2006

Movei 0.08.373   - Arasan 9.2 CCRL            13.5 - 5.5    +12/-4/=3    71.05%
Movei 0.08.373   - E.T.Chess 18.11.05         6.5 - 12.5    +4/-10/=5    34.21%
Movei 0.08.373   - Naum 2.0 (32 bit)          5.0 - 14.0    +2/-11/=6    26.32%
Movei 0.08.373   - Petir 3.99d                9.5 - 9.5    +6/-6/=7    50.00%
Movei 0.08.373   - Pseudo 0.7c                7.0 - 12.0    +5/-10/=4    36.84%
Movei 0.08.373   - Scorpio 1.8.1 (32 bit)     4.0 - 15.0    +2/-13/=4    21.05%
Movei 0.08.373   - Scorpio 1.8.2 (32 bit)     6.5 - 12.5    +5/-11/=3    34.21%
Movei 0.08.373   - The Baron 1.7.0            9.0 - 10.0    +6/-7/=6    47.37%
Movei 0.08.373   - Thinker 4.7a               5.0 - 14.0    +3/-12/=4    26.32%
Movei 0.08.373   - WildCat 6                  12.0 - 7.0    +10/-5/=4    63.16%


Thanks for your results.

Note that I worked on the program and I added parameters and during adding them found something that was not my intention so I changed it.

I am not sure if the change made it better or worse but first test results by me were encouraging.

Movei beated Hermann 16.5-3.5 in FRC at 1+1 and it never got results that was so good earlier in the same 20 games.

I guess that with more games result is going to be worse but I decided to.
stop the match and send the version to Dann Corbit and he promised to send me later today a binary so I hope that 00_8_378 can be the next version to be tested.

Uri
Uri Blass
 
Posts: 40
Joined: Tue Jun 20, 2006 12:44 pm


Return to CCRL Public

Who is online

Users browsing this forum: Google [Bot] and 4 guests