frankp wrote: ↑Wed Aug 01, 2018 4:36 pm
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.
Seems an identical leela (LC0) net, but trained differently or on different data - I think.
Rather like a clone of, say, SF but with different pst.
Interesting to see how it performs, particularly if the (identical) leela net is trained on (a relatively) small set of only human games.
(Raises interesting questions about what the GPL licence applies to, should Albert decide to sell it as the interview linked on leela forum suggests may be a possibility. But that is a different subject.).
frankp wrote: ↑Wed Aug 01, 2018 4:36 pm
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'.
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
Thanks, I didn't know you tried supervised learning as sanity check. How was it performing at tactics? One of the things which bother me is the dismal tactical abilities of all the attempts so far. And the regression in these with the new nets, especially in the main branch. Perhaps sometime in the future one has to abandon the pure UCT MCTS search for some hybrids with AB in the manner Daniel exemplified.
The initial AlphaGo was trained using some server human games, maybe as weak as KGS games (IIRC). But it played at about 1000 Elo points above that level against Lee Sedol (well, on a monster hardware). Chess has compressed Elo differences compared to Go, so supervised learning might give 500 Elo points better results than say training games of averaging 2500 level humans, on a strong hardware. 3000 FIDE Elo points is perhaps achievable with supervised learning (on a strong hardware). Current main- and test- server nets on a strong hardware are considerably stronger than that, say 3200 FIDE Elo level. So, yes, it does seem plausible that Deus X will be weaker in TCEC than Lc0 with a reasonable net of the developers using reinforcement learning. I think 20b nets from testserver scale better to LTC and hardware than 15b mainserver nets, so in TCEC conditions those 20b nets should be preferred, maybe in 10160-10190 range of IDs.
Another issue: it was going so well with 6x64 nets on testserver, 15 times faster games, quick saturation at high Elo value (surely above 3100 CCRL 40/4'). They only scaled badly with TC, but that's understandable. Wasn't it simpler to proceed 6b -> 10b -> 15b -> 20b? I guess even with 10b nets we would have had stronger nets than the current ones, and it was real fast, a matter of days. Then to slower 15b and to slowish and seemingly tricky 20b. I don't even know how one would tune CPUCT, FPU and such, as in my tries their optima strength-wise depend heavily on time control (and hardware) used.
Honestly I have a strong suspicion the tactical weakness of Leela is due to the google algorithm and not due to a weakness of the neural net. A smaller neural net could help a lot in that respect, simply allowing more nodes to be searched at the cost of precision. It is also not clear to me why UCT search should be better than a modified AB search, such as Matthew Lai's probability based search, once the network has been trained, that's actually something I want to try with Leela when I have time, but with my thesis deadline coming up and me enjoying working on Winter I don't think that will happen in the near future.
I believe the original AlphaGo was only bootstrapped with human games. It was not trained in a supervised fashion after that iirc. The games are very different though, it is quite possible that in chess supervised training will work very well, even if it doesn't in Go.
frankp wrote: ↑Wed Aug 01, 2018 4:36 pm
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
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
frankp wrote: ↑Wed Aug 01, 2018 4:36 pm
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).
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
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.
#WeAreAllDraude #JusticeForDraude #RememberDraude #LeptirBigUltra "Those who can't do, clone instead" - Eduard ( A real life friend, not this forum's Eduard )
frankp wrote: ↑Wed Aug 01, 2018 4:36 pm
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'.
frankp wrote: ↑Wed Aug 01, 2018 4:36 pm
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'.
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.
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
Albert Silver wrote: ↑Wed Aug 01, 2018 3:43 pmChanging 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.
