Garbochess v2.20 released

[gladius]

You can get it from http://www.forwardcoding.com/projects/g ... 2-v220.zip.

New in this version:
- Fail low re-search now researches on same ply
- Better search ordering in PV nodes only now
- Removed 1.5 ply LMR
- Increased pawn shelter values
- Better null move verification

It looks to be about a 40 elo increase from v2.11.

[Tony Thomas]

By the time I get about 50 games from a version, you release a new version.

[gladius]

I've had a lot of time to work on the engine lately .

[Graham Banks]

Hi Gary,

Please note that at this rate of releasing new versions you can't expect a sufficient number of games for rating lists. GC 2.02 had played only 24 games for the CCRL blitz list when 2.10 came out. I switched to 2.10 but up to now it has played only 139 games which means an error margin of 50 (!) elo in its strength.
Although I am looking forward to the progress of your promising engine, I decided to stick with version 2.10 until the margin goes down to 30, and only then will I switch to a version that is current at that time.

Hi Gabor,

whilst I can see your point and also find that a little frustrating, I don't think I'd ever try to discourage a keen engine author like Gary.
GarboChess could well be one of the giants of the future, considering the rate at which it's improving.

Regards, Graham.

[gladius]

Hi Gary,

Please note that at this rate of releasing new versions you can't expect a sufficient number of games for rating lists. GC 2.02 had played only 24 games for the CCRL blitz list when 2.10 came out. I switched to 2.10 but up to now it has played only 139 games which means an error margin of 50 (!) elo in its strength.
Although I am looking forward to the progress of your promising engine, I decided to stick with version 2.10 until the margin goes down to 30, and only then will I switch to a version that is current at that time.

I certainly understand! I basically release the engine when there has been a significant strength increase, and I think it's relatively stable/bug free. I don't expect each new version to be tested extensively, and I'm happy to see any results at all .

[swami]

You can get it from http://www.forwardcoding.com/projects/g ... 2-v220.zip.

New in this version:
- Fail low re-search now researches on same ply
- Better search ordering in PV nodes only now
- Removed 1.5 ply LMR
- Increased pawn shelter values
- Better null move verification

It looks to be about a 40 elo increase from v2.11.

Hi Gary,

Please note that at this rate of releasing new versions you can't expect a sufficient number of games for rating lists. GC 2.02 had played only 24 games for the CCRL blitz list when 2.10 came out. I switched to 2.10 but up to now it has played only 139 games which means an error margin of 50 (!) elo in its strength.
Although I am looking forward to the progress of your promising engine, I decided to stick with version 2.10 until the margin goes down to 30, and only then will I switch to a version that is current at that time.

Sorry, Gabor. It is not the author's problem, it is the tester's. Authors can release any number of versions they want, testers should find _the_ way_ to deal with it. Personally I'd like to see as many versions as possible without limit. It's just me, also I suspect there are others like Harun, Tony etc who may share my view but I can't speak for them.

Silencing the other party is something that irks me

[Tony Thomas]

I prefer to get a new version once every 3 months or so, instead of once every weak.. Again, 100 points improvement every 3 month is better than 33 every few weeks.

[swami]

I prefer to get a new version once every 3 months or so, instead of once every weak.. Again, 100 points improvement every 3 month is better than 33 every few weeks.

100 points every 3 months is better than 33 every few weeks? man, what are you upto?! Most would prefer the latter.

[Tony Thomas]

I prefer to get a new version once every 3 months or so, instead of once every weak.. Again, 100 points improvement every 3 month is better than 33 every few weeks.

100 points every 3 months is better than 33 every few weeks? man, what are you upto?! Most would prefer the latter.

At 33 points every few weeks, the program would be slowly crawling up a rating list. I like to see it making a jump. If it is 100 every quarter of the year, after couple of releases it will surely be accused as a clone, and we will have heated debates about weather or not it is an original engine. It is also possible that I drank little too much water, or I messed with goat too much...