Komodo 9.42

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

User avatar
Leto
Posts: 2071
Joined: Thu May 04, 2006 3:40 am
Location: Dune

Re: Komodo 9.42

Post by Leto »

Ozymandias wrote:
Leto wrote:
Ozymandias wrote:
mjlef wrote:Just a reminder that the error bars are pretty big in most rating lists. […] with error margins + and - 19 elo, it is hard to really say who is better.
That goes without saying. But when it's clearly better, like 9.2, it will shine through. Maybe 9.4 will also be noticeably stronger, but in this case, the problem has been bugs. All in all, it seems like 4 updates in less than a year, might be too much of a commitment.
I don't believe 9.2 is better than 9.3. Just because one list has 9.2 above 9.3 doesn't necessarily mean "it's clearly better," especially when CEGT has 9.3 over 9.2 except in the case of 4CPU configurations due to the bug that 9.3 has which only affects it when running on 4 threads. Considering the error bars there's a good chance 9.3 would move ahead of 9.2 in that list once they add more games.
I don't believe 9.2 is better than 9.3 either, but they're too close to tell apart. Same thing could be said for 9 and 9.1. On the other hand, 9.2 is clearly better than 9.1
I have 9.1 at 26 elo above 9.0 in the 12CPU category at CEGT 40/4, scoring 68.2% against opposition rated 3261, while K9.0 12CPU scored 61.9% against opposition rated 3285.

CEGT 40/20 has K9.1 4CPU 9 elo above K9.0 4CPU.

Perhaps the higher the thread count, the more elo K9.1 gains over K9.0.

As for Komodo 9.42 it appears to be doing quite well in the 40/4 12CPU category, scoring 67% in 120 games against the same opposition K9.3 12CPU faced. K9.3 12CPU scored 63.2%.
whereagles
Posts: 565
Joined: Thu Nov 13, 2014 12:03 pm

Re: Komodo 9.42

Post by whereagles »

Ozymandias wrote:I don't believe 9.2 is better than 9.3 either, but they're too close to tell apart. Same thing could be said for 9 and 9.1. On the other hand, 9.2 is clearly better than 9.1
Stuff like this happens a lot in statistics like post-hoc tests...

A = B
B = C
C > A

where "=" means "statistically indistinguishable" to a certain significance.
shrapnel
Posts: 1339
Joined: Fri Nov 02, 2012 9:43 am
Location: New Delhi, India

Re: Komodo 9.42

Post by shrapnel »

JJJ wrote:Which contempt do you use plz ?
And what kind of game yo dou play online ? Komodo vs Engine player ?
Default Contempt 10 is good enough.
Obviously against Engine players, humans are weak and boring.
i7 5960X @ 4.1 Ghz, 64 GB G.Skill RipJaws RAM, Twin Asus ROG Strix OC 11 GB Geforce 2080 Tis
User avatar
Ozymandias
Posts: 1537
Joined: Sun Oct 25, 2009 2:30 am

Re: Komodo 9.42

Post by Ozymandias »

Leto wrote:I have 9.1 at 26 elo above 9.0 in the 12CPU category at CEGT 40/4

CEGT 40/20 has K9.1 4CPU 9 elo above K9.0 4CPU.
There aren't any 12 CPU tests around to compare.

CCRL 40/40 has K9.1 4CPU only 3 ELO above K9 4CPU. Between the two of them, I'd say those versions are too close for comfort.
peter
Posts: 3412
Joined: Sat Feb 16, 2008 7:38 am
Full name: Peter Martan

Re: Komodo 9.42

Post by peter »

For Deep Shredder 12 it works too, but not for Hiarcs CE, because for the latter one "Clear Hash"- command seems not to do what it should with komodo
Peter.
User avatar
Leto
Posts: 2071
Joined: Thu May 04, 2006 3:40 am
Location: Dune

Re: Komodo 9.42

Post by Leto »

Leto wrote:

As for Komodo 9.42 it appears to be doing quite well in the 40/4 12CPU category, scoring 67% in 120 games against the same opposition K9.3 12CPU faced. K9.3 12CPU scored 63.2%.
Now 64.4% after 243 games.
shrapnel
Posts: 1339
Joined: Fri Nov 02, 2012 9:43 am
Location: New Delhi, India

Re: Komodo 9.42

Post by shrapnel »

I must say that introducing the new "Contempt for White" parameter has been really good for me, at least.
Previously, I used to be held to draws very often with ppl using weaker hardware.
This is no longer the case, in the sort of games I play, where it is possible to change Settings at will, even in mid-game .
i7 5960X @ 4.1 Ghz, 64 GB G.Skill RipJaws RAM, Twin Asus ROG Strix OC 11 GB Geforce 2080 Tis
User avatar
Leto
Posts: 2071
Joined: Thu May 04, 2006 3:40 am
Location: Dune

Re: Komodo 9.42

Post by Leto »

390 games completed with K9.42 12CPU, scoring 65% against opposition rated about 100 elo below K9.3 12CPU. It's the same score I got with 110 games with the 9.40 version, so I will merge them, making a total of 500 games thus far.
mjlef
Posts: 1494
Joined: Thu Mar 30, 2006 2:08 pm

Re: Komodo 9.42

Post by mjlef »

peter wrote:Hi Mark!
Found a workaround for Persistent Hash at least for Arena 3.0:
If you give "Save Hash"- command in Arena, you can reload it if you give "Clear Hash" command before "Load Hash".
Having done so, you can even reload it in Fritz14, but also only if hash is empty, just after loading the engine or having given "Clear Hash"- command first.
Still no way to store directly in Fritz- GUI.
Funny, isn't it?
:)
I think most of the issues revolve around when the GUI decides to actually send a command, or commands to the engine. I have found that opening the parameters part of the GUI. Editing the Hash File Name. Then closing the parameters menu forces the GUI to send the new hash name to the engine. Then open the parameters again, then click Save Hash to File, then closing the parameters menu again. Later, you can open it and Load Hash from File.

It seems many GUIS wait until the parameters menu is closed to send things. It may be that in some of your GUIs, clicking Clear Hash force the GUI to send the other commands. There is a log feature in Komodo that lets you at least see what the engine sees from the GUI.

Mark
peter
Posts: 3412
Joined: Sat Feb 16, 2008 7:38 am
Full name: Peter Martan

Re: Komodo 9.42

Post by peter »

Did you try the Clear Hash- command before the one for Load Hash?

If stored in Arena or Shredder you can easily reload after emptying the hash before giving this second one command and in this way that (reloading) works in Fritz too.

There's just no way to store in Fritz, but it works fine in Shredder and Arena, and to reload in any GUI (except Hiarcs CE, there Clear Hash doesn't work at all and switching to another engine in between doesn't help neither because the Load Hash isn't active before komodo hasn't started computing first) you just have to give Clear before Load.

That's the whole problem, Mark, and now that you know so, I'm sure you'll manage to make things work a little more less complicated, if you're really interested in it.
If not, I don't care anymore neither, if I now and then really want to use this feature of komodo's, I do now know how anyhow
:)
Peter.