Jumbo 0.4.0 64-bit Gauntlet for CCRL 40/40

[Graham Banks]

PGN - http://kirill-kryukov.com/chess/discuss ... p?id=38854

[zenpawn]

Some impressive "upsets." Congratulations to Jumbo's author.

[Guenther]

PGN - http://kirill-kryukov.com/chess/discuss ... p?id=38854

Something is wrong with the download. It is only 1 KB extracted and contains no games.

[Graham Banks]

Something is wrong with the download. It is only 1 KB extracted and contains no games.

Ooops!

Here is the correct link:
http://kirill-kryukov.com/chess/discuss ... p?id=38855

[Guenther]

Something is wrong with the download. It is only 1 KB extracted and contains no games.

Ooops!

Here is the correct link:
http://kirill-kryukov.com/chess/discuss ... p?id=38855

Ok, thanks :)

There seems to be a problem with Amundsens time management in this gauntlet?
I see it always uses only 25-30% of the available time?
(25% would be 2 doublings for the opponent)
Did this also happen for other versions?

Actually the strange 8:0 vs. Amundsen was exactly the reason I wanted to check the PGN...
OTH I don't remember how strong this version was, but there is definetely
sth buggy regarding tc management.

Guenther

[Graham Banks]

Something is wrong with the download. It is only 1 KB extracted and contains no games.

Ooops!

Here is the correct link:
http://kirill-kryukov.com/chess/discuss ... p?id=38855

Ok, thanks

There seems to be a problem with Amundsens time management in this gauntlet?
I see it always uses only 25-30% of the available time?
(25% would be 2 doublings for the opponent)
Did this also happen for other versions?

Actually the strange 8:0 vs. Amundsen was exactly the reason I wanted to check the PGN...

Guenther

Not sure Guenther.
I do know that Amundsen was the lowest rated opponent in the gauntlet.

[Sven]

Something is wrong with the download. It is only 1 KB extracted and contains no games.

Ooops!

Here is the correct link:
http://kirill-kryukov.com/chess/discuss ... p?id=38855

Ok, thanks

There seems to be a problem with Amundsens time management in this gauntlet?
I see it always uses only 25-30% of the available time?
(25% would be 2 doublings for the opponent)
Did this also happen for other versions?

Actually the strange 8:0 vs. Amundsen was exactly the reason I wanted to check the PGN...
OTH I don't remember how strong this version was, but there is definetely
sth buggy regarding tc management.

Guenther

I would not call 8:0 "strange". 80:0 would be. 8:0 happens sometimes, even for a rating difference of only about 100 Elo points. Note that Jumbo also lost 0:6 vs. Patzer and Comet. It is normal business that most results are within the expected range based on rating differences but some results aren't. Think of throwing dice: a "5" or "6" eight times in a row is very rare - but it happens. Just not too often (on average in one of 6561 series of throwing dice eight times).

If Amundsen 0.65.1 has time management problems then this is one of the reasons why it played weaker than Jumbo in that gauntlet.

Also the Elo rating is kind of an average. Against some opponents results will be better than expected, even in a high number of games, against some opponents they will be worse, and against the majority they will roughly be as expected. It is certainly possible that Jumbo will always perform better than expected against Amundsen 0.65.1 for an Elo difference of currently +89. I played 100 fast games (40/0:03) with cutechess-cli, with 50 different openings. Result:

Code: Select all

Score of Jumbo_0.4.0 vs Amundsen_0.65.1: 67 - 19 - 14  [0.740] 100
ELO difference: 181.70 +/- 71.81

Result of the first 5 games:

Code: Select all

Score of Jumbo_0.4.0 vs Amundsen_0.65.1: 4 - 1 - 0  [0.800] 5

[Guenther]

Something is wrong with the download. It is only 1 KB extracted and contains no games.

Ooops!

Here is the correct link:
http://kirill-kryukov.com/chess/discuss ... p?id=38855

Ok, thanks :)

There seems to be a problem with Amundsens time management in this gauntlet?
I see it always uses only 25-30% of the available time?
(25% would be 2 doublings for the opponent)
Did this also happen for other versions?

Actually the strange 8:0 vs. Amundsen was exactly the reason I wanted to check the PGN...
OTH I don't remember how strong this version was, but there is definetely
sth buggy regarding tc management.

Guenther

I would not call 8:0 "strange". 80:0 would be. 8:0 happens sometimes, even for a rating difference of only about 100 Elo points. Note that Jumbo also lost 0:6 vs. Patzer and Comet. It is normal business that most results are within the expected range based on rating differences but some results aren't. Think of throwing dice: a "5" or "6" eight times in a row is very rare - but it happens. Just not too often (on average in one of 6561 series of throwing dice eight times).

If Amundsen 0.65.1 has time management problems then this is one of the reasons why it played weaker than Jumbo in that gauntlet.

Also the Elo rating is kind of an average. Against some opponents results will be better than expected, even in a high number of games, against some opponents they will be worse, and against the majority they will roughly be as expected. It is certainly possible that Jumbo will always perform better than expected against Amundsen 0.65.1 for an Elo difference of currently +89. I played 100 fast games (40/0:03) with cutechess-cli, with 50 different openings. Result:

Code: Select all

Score of Jumbo_0.4.0 vs Amundsen_0.65.1: 67 - 19 - 14  [0.740] 100
ELO difference: 181.70 +/- 71.81

Result of the first 5 games:

Code: Select all

Score of Jumbo_0.4.0 vs Amundsen_0.65.1: 4 - 1 - 0  [0.800] 5

Well my question was, if that version of Amundsen _always_ showed that behaviour. Alone the discrepancy in move times vs. Jumbo are equivalent to 120 rating points difference
and it is only 90 pts lower than Jumbo 0.40 in the CCRL rating list, probably despite the 0:8 games included.
Of course it could also be that Jim broke something in that compile, because I checked version 0.60 here and it does _not_ show that behaviour.

[Guenther]

I checked all Amundsen 0.651 games now in 40/40 CCRl and it seems it always had a buggy time management at least for moves/times.
A pity though that I needed to deduct this from the time the opponents used, as there is never the real time control (the adapted one) in the pgn header...

It's also a pity that we still have no GUI which can show the accumulated time, which could make the problem visible in a second.
(but w/o the real tc it would be not as good, as possible of course)