Houdini Versions

[brownsmith89]

maybe you should release a 1.01b version with just the bug fixes, so we can do a proper comparison?

1.01 was crashing in my program.

i enjoy using houdini. i hope it's not too much trouble.

[Robert Houdart]

To reduce the likelyhood of 1.01 crashing, you can increase the hash table size and limit your tests to at most 2 threads.

Robert

[Taner Altinsoy]

They are on the way

[Taner Altinsoy]

Ok my grand tournament finished at last. Total of 3750 games played among Houdini 1.0, 1.01 and 1.02 on a Celeron 1400 laptop with 750 MB ram on a 32 bit WIndows XP. Time control was 30 secs. Each engines played 1250 games with each other.

Program Elo + - Games Score Av.Op. Draws

1 Houdini_w32_1CPU [1.01] : 2805 9 9 2500 51.2 % 2797 55.0 %
2 Houdini_w32_1CPU(1.02) : 2804 9 9 2500 50.8 % 2798 55.4 %
3 Houdini_w32_1CPU : 2791 9 9 2500 48.0 % 2805 55.4 %

I have to apologise to Robert Houdart for the false alarm I caused in the community.

regards,
Taner Altinsoy

[Robert Houdart]

Thank you, Taner, there's absolutely no need to apologize. I'm pleased that your results are quite in line with the announcements I made at the release of the different versions.

Your results clearly demonstrate that a high number of games is required to measure small strength differences. Engine developers are very much aware of the random variability of match results - we see it every day when we test improvements. A change can appear very promising after 500 games, to end up completely disappointing after 3000 games.

Robert

[halestorm]

What time controls do you like to see your engine tested at, Robert? I understand the statistical reasons for thousands, not hundreds. Do you like 1+0, 1+1? Something more? What would benefit you, and against what opponent?

[Robert Houdart]

I prefer time controls with increments. Without increment you're basically crippling all endgame play - at the end neither engine will have the time for playing any decent game.
So 1+1 is definitely more interesting than 1+0.

Robert

[Razor]

Robert,

Would it be possible to include some parameters for the user to change that would effectively cause the engine to favour a more aggressive/attacking style?

BTW, in my opinion H102 is as strong as R4 (any performance difference between these two engines are within the calculation noise limits you would expect so well done - great job. As far as reliance on the engine not throwing an exception error then H102 is streets ahead of R4 - I get a very reliable stream of data from H102 on an i7 quad (even after a couple of days of analysis) whereas R4 I have to watch all the time, if I go beyond an hour I'm luck to escape without the engine freezing which leads to the user just not trusting the analysis. I would also like to comment on the evaluation scoring - in my opinion H102 is much more accurate than DS12, H13.1, R4, and S1.8 - they are all usually displaying higher scores (usually much more optimistic) - when H102 scores high it is all over - I'm sure there is still plenty of work to do in this area; just thought I would pass back these comments as they are in my opinion well deserved.

[Robert Houdart]

Would it be possible to include some parameters for the user to change that would effectively cause the engine to favour a more aggressive/attacking style?

It's true that Houdini's style currently is not very aggressive, I'll have to study what can be tuned to make it a little more Tal-like than Karpov-like without overwhelming the end-user with zillions of UCI options, and without losing too much strength.
Houdini 1.03 is on schedule for July 15, I'll consider your request for the version after 1.03.

As far as reliance on the engine not throwing an exception error then H102 is streets ahead of R4 - I get a very reliable stream of data from H102 on an i7 quad (even after a couple of days of analysis) whereas R4 I have to watch all the time, if I go beyond an hour I'm luck to escape without the engine freezing which leads to the user just not trusting the analysis.

Thanks for your kind words. Version 1.02 appears to be very stable indeed, I haven't witnessed a single crash so far.

Robert

[royb]

Would it be possible to include some parameters for the user to change that would effectively cause the engine to favour a more aggressive/attacking style?

It's true that Houdini's style currently is not very aggressive, I'll have to study what can be tuned to make it a little more Tal-like than Karpov-like without overwhelming the end-user with zillions of UCI options, and without losing too much strength.
Houdini 1.03 is on schedule for July 15, I'll consider your request for the version after 1.03.

As far as reliance on the engine not throwing an exception error then H102 is streets ahead of R4 - I get a very reliable stream of data from H102 on an i7 quad (even after a couple of days of analysis) whereas R4 I have to watch all the time, if I go beyond an hour I'm luck to escape without the engine freezing which leads to the user just not trusting the analysis.

Thanks for your kind words. Version 1.02 appears to be very stable indeed, I haven't witnessed a single crash so far.

Robert

Will Houdini 1.3 offer a 64-bit Linux compile as well? I'd really like to have that!!