On-line engine blitz tourney July

[Joost Buijs]

a word of caution: last time I censored (per advice in chat) micromaxXb to avoid the annoying shouts it made each move,
yesterday this caused the Cheng-umax game not starting automatically, delaying the tournament.
then I messed up the manual match without specifying time control so game was aborted and adjourned so the censor adventure
cost me a half-point. I suggest anyone who also censored u-max in the past to uncensor in order to avoid similar problems

It was not my decision to adjudicate the Cheng-umax game a draw, I would have given Cheng the full point because the chance of umax getting a draw against Cheng is almost zero. Of course, Lisebeth (eendje) doesn't know this.

[mar]

a word of caution: last time I censored (per advice in chat) micromaxXb to avoid the annoying shouts it made each move,
yesterday this caused the Cheng-umax game not starting automatically, delaying the tournament.
then I messed up the manual match without specifying time control so game was aborted and adjourned so the censor adventure
cost me a half-point. I suggest anyone who also censored u-max in the past to uncensor in order to avoid similar problems

It was not my decision to adjudicate the Cheng-umax game a draw, I would have given Cheng the full point because the chance of umax getting a draw against Cheng is almost zero. Of course, Lisebeth (eendje) doesn't know this.

no that's ok, I messed up and forgot to match with correct tc.
I'm aware that multiple people censored umaxXb in the previous event so the same thing might
happen in the future.

[Joost Buijs]

I've been working a lot on Vice the last few months - fixing the bugs, adding ponder and quite a few search features / lots of testing, but wow, all the other engines are so strong. From my testing, Vice is about 400 elo better than before, nevertheless, no chance here!!

Even Euwe, one place above Vice, is 2800+ on CCRL.

It's amazing how good people are at writing these things, there are so many really strong programs. Nevertheless, watching the games remains really fascinating,

It's nice to play in the same tournament as Rebel - I can remember printing the "How Rebel Plays Chess" web pages in the early 2000s, and reading it again and again.

Hopefully there will be some more engines < 2400 at some point

Thanks again,

Richard

In the past there were more engines with ratings < 2400, but they all got stronger and stronger. HGM had several < 2400 engines running, but I haven't seen them for a long time.

I don't know how big the difference between CCRL ratings and ratings on our server is, I estimate about 600 points. Compared to human ratings our ratings are too low and CCRL ratings are too high.

I've been thinking about increasing the ratings on the server by 300 points, maybe I will do this some day.

[Rebel]

Is it possible to download the PGN of all games played ?

No, I dont think this is possible because many games played are not available for download, and I don't see what this has to do with training a neural-network, you don't need an infinite number of games to train a neural-net. Of course more games are usually better, but above a certain threshold it doesn't get better anymore.

A network trains on sub-pattterns that occur in positions, many positions contain the same sub-patterns, it's important that the positions you train on are labeled correctly, e.g. +1 for a win, 0 for a draw and -1 for a loss. When you use a weak program to produce the games you derive these positions from, the labels will be way off, and you will get a weak network.

My question has nothing to do with NNUE programming but is about book preparation, fixing holes. Instead of manually downloading every game played in the monthly tournaments resulting in hundreds of emails, I am asking you to provide the PGN of the monthly tournaments. I am pretty sure you have that.

[Joost Buijs]

Is it possible to download the PGN of all games played ?

No, I dont think this is possible because many games played are not available for download, and I don't see what this has to do with training a neural-network, you don't need an infinite number of games to train a neural-net. Of course more games are usually better, but above a certain threshold it doesn't get better anymore.

A network trains on sub-pattterns that occur in positions, many positions contain the same sub-patterns, it's important that the positions you train on are labeled correctly, e.g. +1 for a win, 0 for a draw and -1 for a loss. When you use a weak program to produce the games you derive these positions from, the labels will be way off, and you will get a weak network.

My question has nothing to do with NNUE programming but is about book preparation, fixing holes. Instead of manually downloading every game played in the monthly tournaments resulting in hundreds of emails, I am asking you to provide the PGN of the monthly tournaments. I am pretty sure you have that.

Sorry, I misunderstood what you meant.

The tourney manager 'mamer' doesn't store the games but relies upon the underlying server software. The server stores the last 50 games games of a player in it's player directory, but they are not in PGN-format. To get all the games of a tournament in PGN-format is quite a lot of work because this is not supported by the very old Lasker server software. Of course it's possible to write a function to translate all the games of a tournament to PGN and mail them in a single packet, but I don't want to spend time on doing this.

The server has a function to convert a single game to PGN and send it by email, but this is buggy and often produces a PGN which contains a bad move, so this is not very usable too.