Open Source Blitz Rating List: IvanHoe

[lucasart]

Hallelujah!
I managed to compile IvanHoe 999946h on Linux, using gcc 4.6 and with link time optimizations (which aren't used in the klo compile). So I can safely say I have the best IvanHoe Linux compile now

Remember I told you KLO's were the only Ivanhoe Linux compiles. I never said they were worth a shit.

george

actually on linux it's really easy to compile programs yourself. at least i know what i'm using. i can't wait for my computer to be available (when Pepito 1.59 test is finished), and let IvanHoe 999946h kick some asses in my rating list

Look, far be it from me to tell you how to best handle your testing, but I think you are not going to get the best of comparisons from version to version if you are comparing any of your compile results to his compile results. IOW, I would not compare your compile against Stockfish to his compile against Stockfish. Then when one is better you don't know if it is the version or the compile that made the difference. You might end up with too many variables.

george

no, that's NOT the way to compare. the point is that the compiles are functionnally identical. so all you need is to run something like

Code: Select all

uci
ucinewgame
position startpos
go depth 20

and see what the time in ms is. using real games is stupid because to measure such a difference reliably, you have to do a sequential probability ratio test, and it will likely take thousands and thousands of games !!

of course i'll take my compile of 46a and his compile of 46a, using the exact same source code.

[geots]

Hallelujah!
I managed to compile IvanHoe 999946h on Linux, using gcc 4.6 and with link time optimizations (which aren't used in the klo compile). So I can safely say I have the best IvanHoe Linux compile now

Remember I told you KLO's were the only Ivanhoe Linux compiles. I never said they were worth a shit.

george

actually on linux it's really easy to compile programs yourself. at least i know what i'm using. i can't wait for my computer to be available (when Pepito 1.59 test is finished), and let IvanHoe 999946h kick some asses in my rating list

Look, far be it from me to tell you how to best handle your testing, but I think you are not going to get the best of comparisons from version to version if you are comparing any of your compile results to his compile results. IOW, I would not compare your compile against Stockfish to his compile against Stockfish. Then when one is better you don't know if it is the version or the compile that made the difference. You might end up with too many variables.

george

no, that's NOT the way to compare. the point is that the compiles are functionnally identical. so all you need is to run something like

Code: Select all

uci
ucinewgame
position startpos
go depth 20

and see what the time in ms is. using real games is stupid because to measure such a difference reliably, you have to do a sequential probability ratio test, and it will likely take thousands and thousands of games !!

of course i'll take my compile of 46a and his compile of 46a, using the exact same source code.

I stand corrected. And you are right of course. The only reason I don't like start positions is because testing has a lot of data that has to be kept up with and organized and that is not the fun part. The fun part to me is the actual games played between engines- when you really don't know how a match or RR might turn out.

Chess is a hobby that is for fun, and if the actual engine game playing were removed from the mix, nothing about it would I any longer enjoy. When the enjoyment is no longer there, the next step is burn-out, followed by good-bye.

But you enjoy your method, and I am all for you!


Best regards,

george

[Pablo Vazquez]

Hallelujah!
I managed to compile IvanHoe 999946h on Linux, using gcc 4.6 and with link time optimizations (which aren't used in the klo compile). So I can safely say I have the best IvanHoe Linux compile now

Remember I told you KLO's were the only Ivanhoe Linux compiles. I never said they were worth a shit.

george

actually on linux it's really easy to compile programs yourself. at least i know what i'm using. i can't wait for my computer to be available (when Pepito 1.59 test is finished), and let IvanHoe 999946h kick some asses in my rating list

Look, far be it from me to tell you how to best handle your testing, but I think you are not going to get the best of comparisons from version to version if you are comparing any of your compile results to his compile results. IOW, I would not compare your compile against Stockfish to his compile against Stockfish. Then when one is better you don't know if it is the version or the compile that made the difference. You might end up with too many variables.

george

no, that's NOT the way to compare. the point is that the compiles are functionnally identical. so all you need is to run something like

Code: Select all

uci
ucinewgame
position startpos
go depth 20

and see what the time in ms is. using real games is stupid because to measure such a difference reliably, you have to do a sequential probability ratio test, and it will likely take thousands and thousands of games !!

of course i'll take my compile of 46a and his compile of 46a, using the exact same source code.

I think the best you can do is to compile it yourself from the original source. I just downloaded it from the official page to try to compile it and it was quite easy. It comes with the makefile so I just had to type make.

All the compiles out there are probably the same within 5% of speed difference. They are just classified as better or worse by playing 50 game matches or even less.

Also, I don't know if there is any functional difference in all the 999946 versions because in the change log it only mentions chess960 and bug fixes. You can probably check that using your method above and checking the node counts.