Engine 40x(2) OVERHAULED!?

[geots]

As the header tells you- significant changes to the engine that just finished the 1000-game match with Houdini 2.0c and a new name it loads with.

40x2 Good For You-and it has begun 100-game matches against both Strelka 5.5 and Houdini 2.0c.

My question was posed to the programmer- could I please run his "MAIN" engine? I was told to think of it in terms of "pay-per-view"- that anyone who sees it in action buys a ticket first.

I asked how much of a strength increase to expect here and told he had no clue- hasn't been tested. That is what I am here for. I considered then just sending the engine back to the "programmer-by-proxy", but changed my mind.

Updates depend on how fast the matches run- as they are both 100-game matches at 5'+5".


Where might Jesus be-

george

[Ajedrecista]

Hello George:

As the header tells you- significant changes to the engine that just finished the 1000-game match with Houdini 2.0c and a new name it loads with.

40x2 Good For You-and it has begun 100-game matches against both Strelka 5.5 and Houdini 2.0c.

My question was posed to the programmer- could I please run his "MAIN" engine? I was told to think of it in terms of "pay-per-view"- that anyone who sees it in action buys a ticket first.

I asked how much of a strength increase to expect here and told he had no clue- hasn't been tested. That is what I am here for. I considered then just sending the engine back to the "programmer-by-proxy", but changed my mind.

Updates depend on how fast the matches run- as they are both 100-game matches at 5'+5".


Where might Jesus be-

george

It sounds interesting. 100 games at 5' + 5"/move should bring error bars over ± 50 Elo, much bigger than the expected Elo difference itself... although I suppose that this is better than nothing. The 1000-game match was just too good.

With similar draw ratios that in the 1000-game match, Houdini should win around 61 - 39 for getting a LOS of around 99.9% (as it reached in that match), so it is possible but very unlikely IMHO. The importance in the number of games is more important that I thought at a first glance, and this is a thing that many top programmers know. Anyway, it will be a good enjoyment.

Since I have no interest in keep private the improvements I made in my two simple programmes, I have uploaded them:

LOS_and_Elo_uncertainties_calculator_&_Minimum_score_for_no_regression.rar (0.62 MB)

These links will not be active forever. The results can be posted anywhere... but you always must be careful, because I do not offer a warranty of the rightness of the results, although most of them seem correct from my POV. Both programmes are open source, so you may laugh with my limited programming skills and lack of optimizations!

Regards from Spain.

Ajedrecista.

[geots]

Hello George:

As the header tells you- significant changes to the engine that just finished the 1000-game match with Houdini 2.0c and a new name it loads with.

40x2 Good For You-and it has begun 100-game matches against both Strelka 5.5 and Houdini 2.0c.

My question was posed to the programmer- could I please run his "MAIN" engine? I was told to think of it in terms of "pay-per-view"- that anyone who sees it in action buys a ticket first.

I asked how much of a strength increase to expect here and told he had no clue- hasn't been tested. That is what I am here for. I considered then just sending the engine back to the "programmer-by-proxy", but changed my mind.

Updates depend on how fast the matches run- as they are both 100-game matches at 5'+5".


Where might Jesus be-

george

It sounds interesting. 100 games at 5' + 5"/move should bring error bars over ± 50 Elo, much bigger than the expected Elo difference itself... although I suppose that this is better than nothing. The 1000-game match was just too good.

With similar draw ratios that in the 1000-game match, Houdini should win around 61 - 39 for getting a LOS of around 99.9% (as it reached in that match), so it is possible but very unlikely IMHO. The importance in the number of games is more important that I thought at a first glance, and this is a thing that many top programmers know. Anyway, it will be a good enjoyment.

Since I have no interest in keep private the improvements I made in my two simple programmes, I have uploaded them:

LOS_and_Elo_uncertainties_calculator_&_Minimum_score_for_no_regression.rar (0.62 MB)

These links will not be active forever. The results can be posted anywhere... but you always must be careful, because I do not offer a warranty of the rightness of the results, although most of them seem correct from my POV. Both programmes are open source, so you may laugh with my limited programming skills and lack of optimizations!

Regards from Spain.

Ajedrecista.

Jesus- 2 important points I need to make:

1. I might do a lot of things, but laughing at any of your work would never be one of them. That you can take to the bank.

2. I understand what you are saying. The 1000 game match sort of spoiled both you and me. We want to see a lot more of it. There may be some- but surely not a LOT more. It just can't be done. The results would be great- but they would be too few. We can think of a 50 game match or a 100 game match as still being a lot more reliable number of games than is played for any "world championship" between engines. I can give the public results from 50 or 100 games and let them make up their own minds about what it means. Sometimes you just do what you can.


Best,

gts