TCEC season 13, 2 NN engines will be participating, Leela and Deus X

[Albert Silver]

Albert, what is your source of games to train the Leela network with?
Just interested, since Leela is now trained against >20M games and this has taken months of around 200 contributors.

Megabase 2018

[Albert Silver]

Albert, what is your source of games to train the Leela network with?
Just interested, since Leela is now trained against >20M games and this has taken months of around 200 contributors.

I would imagine one of the primary reasons he is using "Human" games (I thought I read somewhere he is using correspondence games? If so, hardly human imo) is precisely because it uses several orders of magnitude less resources. Most of the resources required from those contributors were to create self play games, sometimes between absolute garbage quality networks. Each network is trained on a subset of those games, not all of them. If you do not have insane resources, like a distributed project or large company can have, then going the zero approach is not really feasible at the moment.

On the other hand I am not convinced you need too many games in order to train a network capable of producing much higher quality games than your dataset.

No, the reason I'm using human games is because that was the idea from the beginning. To see how far the neural network could go from purely human games. Please note that the structure of this particular neural network is about 4 times smaller than the one submitted to TCEC by the Leela team (10161 I think).

[AndrewGrant]

Note that the things I have posted here are simply speculative. Although I am involved in the TCEC games, I am not actually in-the-know about the inner-workings of Anton, TCEC, and others.

It has been pointed out to me that some take my writings to be insulting. This is not my intend, and in fact, far from it.

I do my best to hold engines, authors, and whomever else to a high standard. When I see something I find odd, I question it.

It seems solidified that DeusX will play in TCEC. In this case, I wish Albert the best of luck. The same to the Leela team. May the best Net win.

[frankp]

Albert, what is your source of games to train the Leela network with?
Just interested, since Leela is now trained against >20M games and this has taken months of around 200 contributors.

I would imagine one of the primary reasons he is using "Human" games (I thought I read somewhere he is using correspondence games? If so, hardly human imo) is precisely because it uses several orders of magnitude less resources. Most of the resources required from those contributors were to create self play games, sometimes between absolute garbage quality networks. Each network is trained on a subset of those games, not all of them. If you do not have insane resources, like a distributed project or large company can have, then going the zero approach is not really feasible at the moment.

On the other hand I am not convinced you need too many games in order to train a network capable of producing much higher quality games than your dataset.

Thanks.
I found some information on the leela forum site.

If you mean Chris Whittington's fountain of wisdom, I would question the use of the word 'information'.

I was referring to the following links to your interview and a a statement from the leela team.
http://www.chessdom.com/statements-by-d ... o-authors/
https://www.youtube.com/watch?v=CpjvvcfbdR4

EDIT: just read the leela forum. No I was not referencing this, rather info provided through the leela chat. Forum was evidently the wrong word.

[Albert Silver]

Albert, what is your source of games to train the Leela network with?
Just interested, since Leela is now trained against >20M games and this has taken months of around 200 contributors.

I would imagine one of the primary reasons he is using "Human" games (I thought I read somewhere he is using correspondence games? If so, hardly human imo) is precisely because it uses several orders of magnitude less resources. Most of the resources required from those contributors were to create self play games, sometimes between absolute garbage quality networks. Each network is trained on a subset of those games, not all of them. If you do not have insane resources, like a distributed project or large company can have, then going the zero approach is not really feasible at the moment.

On the other hand I am not convinced you need too many games in order to train a network capable of producing much higher quality games than your dataset.

Thanks.
I found some information on the leela forum site.

If you mean Chris Whittington's fountain of wisdom, I would question the use of the word 'information'.

I was referring to the following links to your interview and a a statement from the leela team.
http://www.chessdom.com/statements-by-d ... o-authors/
https://www.youtube.com/watch?v=CpjvvcfbdR4

EDIT: just read the leela forum. No I was not referencing this, rather info provided through the leela chat. Forum was evidently the wrong word.

It is easy to think of the NN as just a very elaborate evaluation function, but it goes quite a bit deeper than that, and is the reason that some short-term moves, however forced, are permanently out of the reach of the NN whatever the depth or time spent. This is by no means to diminish the need and importance of a solid executable to run the MCTS and use the NN, but even tactics depend very much on the quality of the NN and how it was developed. The dataset I used was the same for over 8 tries, and some 2000 hours of computer time in all, yet 7 of those tries bombed, some quite badly. I invested a lot of time, hundreds of hours, testing, learning, preparing the material, reading god knows how many papers to try to find tweaks and improvements, and more. I have no regrets, and had plenty of fun, but it was a ton of work.

[mclane]

Human games are full of mistakes.
How shall a computer learn it of games full of mistakes ?

[Ozymandias]

Changing random numbers, as you suggest, would mean using some NN of Leela, instead of toiling on this for months as I have, building it from scratch, with numerous stalls and restarts along the way. Building a NN isn't that hard, but building a good one, much less a really good one, is very hard.

Do you plan on writing an article for CB, on how to train nets on your own? I would read it.

As for the whole TCEC debate, I don't care much for the "originality" of an engine, I'd rather see playing the likes of Kelly Kinyama's Self-Learning Stockfish, or the mentioned Scorpio hybrid, or the MTCS version of Komodo (or Deus X, of course)... over a new traditional AB engine.

[frankp]

Albert, what is your source of games to train the Leela network with?
Just interested, since Leela is now trained against >20M games and this has taken months of around 200 contributors.

I would imagine one of the primary reasons he is using "Human" games (I thought I read somewhere he is using correspondence games? If so, hardly human imo) is precisely because it uses several orders of magnitude less resources. Most of the resources required from those contributors were to create self play games, sometimes between absolute garbage quality networks. Each network is trained on a subset of those games, not all of them. If you do not have insane resources, like a distributed project or large company can have, then going the zero approach is not really feasible at the moment.

On the other hand I am not convinced you need too many games in order to train a network capable of producing much higher quality games than your dataset.

Thanks.
I found some information on the leela forum site.

If you mean Chris Whittington's fountain of wisdom, I would question the use of the word 'information'.

I was referring to the following links to your interview and a a statement from the leela team.
http://www.chessdom.com/statements-by-d ... o-authors/
https://www.youtube.com/watch?v=CpjvvcfbdR4

EDIT: just read the leela forum. No I was not referencing this, rather info provided through the leela chat. Forum was evidently the wrong word.

It is easy to think of the NN as just a very elaborate evaluation function, but it goes quite a bit deeper than that, and is the reason that some short-term moves, however forced, are permanently out of the reach of the NN whatever the depth or time spent. This is by no means to diminish the need and importance of a solid executable to run the MCTS and use the NN, but even tactics depend very much on the quality of the NN and how it was developed. The dataset I used was the same for over 8 tries, and some 2000 hours of computer time in all, yet 7 of those tries bombed, some quite badly. I invested a lot of time, hundreds of hours, testing, learning, preparing the material, reading god knows how many papers to try to find tweaks and improvements, and more. I have no regrets, and had plenty of fun, but it was a ton of work.

Sort of nature-nurture debate, I guess.
I now think of leela (and indeed chess) as no more than linear algebra.
The clone comment I made referred to my understanding that the network structure and its implementation in software is a 'clone' of the lc0/leela and not modified in any way. My very limited understanding is then that training can do no more than affect the weighting factors of the 'connections' between the 'neurons' of the net. Absolutely key to how the end products plays chess, though that may be. Educating the net is also a process/art in itself, no doubt.

I make no comment of the issue of TECE allowing a close relative of leela into its competition.
And am interested to see how it performs now you have clarified it is trained on human games. Good luck.

Having said that, the zero, self-play approach is of most interest to me, since if this approach can produce a chess super-intelligence (so to speak) then it cannot be blinkered by training based on human constructs - perhaps. And without trying we will never know.

It would be fascinating to produce a NN based on Carlsen games etc (or club players), but I guess there will never be enough games..... Fascinating area to explore.

[Robert Pope]

Human games are full of mistakes.
How shall a computer learn it of games full of mistakes ?

Lc0 training games are also full of mistakes, on purpose. The fact that one side played a mistake, and lost, is still information that can be leveraged.

It would be of more concern, if the same mistake always got made. E.g. if theory on KID is fundamentally flawed and correct theory never gets played in the database, then that could potentially hamstring the ability to learn the correct theory. This is one of the points of the unsupervised learning - it tries all sorts of crazy things in the games, and sometimes stumbles across a nugget as a result. The downside is that the training data is so much noisier because a good move will often be followed by a blunder, so it requires more data to train on.

[whereagles]

Aren't you guys being a bit too strict??? TCEC is for amusement only, and two NN, trained under very different circumstances, add a nice twist to this season. The shared base code situation is obviously borderline on the regulations, but the competition will certainly gain in interest.