New engine releases 2020

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

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Wolfgang
Posts: 345
Joined: Fri May 12, 2006 11:08 pm

Re: New engine releases 2020

Post by Wolfgang » Fri Mar 27, 2020 10:24 am

xr_a_y wrote:
Fri Mar 27, 2020 7:56 am
....

I think it was igel-220 under cutechess. When end-game is near, Igel starts to use too much time, and eventually lost on time.
Ok, I have neither experience with this version nor with cutechess.
Maybe the (very) small bonus of 0,5 seconds causes problems?!
Our list I refer too is played with 3 seconds bonus (5 minutes per game + 3 seconds per move, 5'+3") where time forfeits are very rare.
GUI is Shredder Classic or Arena 3.5.1

Maybe voffka provides a 2.4.1 with ponder enabled so I can check it out :)
Best
Wolfgang
CEGT-Team

voffka
Posts: 114
Joined: Sat Jun 30, 2018 8:58 pm
Location: Ukraine
Full name: Volodymyr Shcherbyna
Contact:

Re: New engine releases 2020

Post by voffka » Fri Mar 27, 2020 3:13 pm

I am currently running 2000 games between Igel and Topple in 60s+0.6 with and without ponder to see if there is an elo degradation. If elo is the same, I will produce 2.4.1 with ponder enabled.

If elo is not the same, well, I will need to dig :)

P.S. No offense of not using Minic for the test, it's just that I already have some Topple games played, so it makes relative elo comparison easier :)

voffka
Posts: 114
Joined: Sat Jun 30, 2018 8:58 pm
Location: Ukraine
Full name: Volodymyr Shcherbyna
Contact:

Re: New engine releases 2020

Post by voffka » Fri Mar 27, 2020 6:16 pm

I've run some ponder tests Wolfgang,

Image

As you can see, Igel is loosing on time just like Vivien said. Basically Igel looses all games against weaker components (Topple is 125 elo weaker). I am not conformable making a ponder release with this known issue.

It is important to note that in my test I am not using adjudication. The moment I enabled adjudication (3 moves, 600 cp) the score of Igel has improved, because it did not enter in endgame, but still, I'd like to fix the root-cause instead of releasing a buggy release.

User avatar
CMCanavessi
Posts: 886
Joined: Thu Dec 28, 2017 3:06 pm
Location: Argentina

Re: New engine releases 2020

Post by CMCanavessi » Fri Mar 27, 2020 8:53 pm

voffka wrote:
Fri Mar 27, 2020 6:16 pm
I've run some ponder tests Wolfgang,

Image

As you can see, Igel is loosing on time just like Vivien said. Basically Igel looses all games against weaker components (Topple is 125 elo weaker). I am not conformable making a ponder release with this known issue.

It is important to note that in my test I am not using adjudication. The moment I enabled adjudication (3 moves, 600 cp) the score of Igel has improved, because it did not enter in endgame, but still, I'd like to fix the root-cause instead of releasing a buggy release.
Weird, what TC are you using. 1m + 1s works ok and doesn't lose on time.
Follow my tournament and some Leela gauntlets live at http://twitch.tv/ccls

voffka
Posts: 114
Joined: Sat Jun 30, 2018 8:58 pm
Location: Ukraine
Full name: Volodymyr Shcherbyna
Contact:

Re: New engine releases 2020

Post by voffka » Fri Mar 27, 2020 9:14 pm

This is 60s+0.6 increment without adjudication by score or by tbs.

Wolfgang
Posts: 345
Joined: Fri May 12, 2006 11:08 pm

Re: New engine releases 2020

Post by Wolfgang » Fri Mar 27, 2020 10:14 pm

voffka wrote:
Fri Mar 27, 2020 6:16 pm
I've run some ponder tests Wolfgang,

Image

As you can see, Igel is loosing on time just like Vivien said. Basically Igel looses all games against weaker components (Topple is 125 elo weaker). I am not conformable making a ponder release with this known issue.

It is important to note that in my test I am not using adjudication. The moment I enabled adjudication (3 moves, 600 cp) the score of Igel has improved, because it did not enter in endgame, but still, I'd like to fix the root-cause instead of releasing a buggy release.
I've checked all the 800 games with 2.1.0 and 5'+3".
Not a single time forfeit under these conditions, i.e. 3 seconds increment and normal GUI. Strength of opponents has no meaning because we played against weaker/equal/stronger opponents as usual.

To keep a long matter short:
In the end it's your decision what to release or not, as well as it is my decision what to test. Or not...
Best
Wolfgang
CEGT-Team

voffka
Posts: 114
Joined: Sat Jun 30, 2018 8:58 pm
Location: Ukraine
Full name: Volodymyr Shcherbyna
Contact:

Re: New engine releases 2020

Post by voffka » Fri Mar 27, 2020 11:02 pm

Hey Wolfgang,
Wolfgang wrote:
Fri Mar 27, 2020 10:14 pm
I've checked all the 800 games with 2.1.0 and 5'+3".
Thanks for a thorough check and the time you spend on this.

I've identified a bug in Igel and applied the fix. It has worked well in my manual test on 3 games against Topple, I am running 1000 games more to confirm with ponder=on in 60+0.6. If it shows good results I will release an Igel 2.4.0 (for windows only) in a separate folder inside the bin archive at https://github.com/vshcherbyna/igel/releases/tag/2.4.0

voffka
Posts: 114
Joined: Sat Jun 30, 2018 8:58 pm
Location: Ukraine
Full name: Volodymyr Shcherbyna
Contact:

Re: New engine releases 2020

Post by voffka » Fri Mar 27, 2020 11:20 pm

Wolfgang,

I see a bit better results of new Igel in ponder mode against Topple:

Image

Please check the folder 'ponder' inside archive at https://github.com/vshcherbyna/igel/rel ... .0.bin.zip

The fix will be a part of mainstream version of Igel in next releases.

Wolfgang
Posts: 345
Joined: Fri May 12, 2006 11:08 pm

Re: New engine releases 2020

Post by Wolfgang » Sat Mar 28, 2020 12:53 am

Thanks very much.
I'll integrate it in my current tournament and report. May take 2 or 3 days to get a considerable number of games
Best
Wolfgang
CEGT-Team

voffka
Posts: 114
Joined: Sat Jun 30, 2018 8:58 pm
Location: Ukraine
Full name: Volodymyr Shcherbyna
Contact:

Re: New engine releases 2020

Post by voffka » Sat Mar 28, 2020 1:18 am

Hey Wolfgang,
Wolfgang wrote:
Sat Mar 28, 2020 12:53 am
Thanks very much.
I'll integrate it in my current tournament and report. May take 2 or 3 days to get a considerable number of games
Thanks! Sounds like a plan :)

I noticed that with ponder=on Igel is stronger by at least 100 elo more against Topple (also with ponder). Don't know if this is a coincidence or not. I may need to test ponder against top 50 engines to see if there is some elo improvement because of my fix.

P.S. In order to avoid spamming this thread, I suggest to move further conversations about Igel to a dedicate thread I created a while ago: viewtopic.php?t=67890

Post Reply