It is currently Thu Mar 28, 2024 2:40 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 418 posts ]  Go to page Previous  1 ... 4, 5, 6, 7, 8, 9, 10 ... 21  Next
Author Message
Offline
 Post subject: Re: LZ's progression
Post #121 Posted: Thu Sep 13, 2018 5:47 am 
Lives in gote

Posts: 502
Liked others: 1
Was liked: 153
Rank: KGS 2k
GD Posts: 100
KGS: Tryss
Quote:
Has LZ also built up a model of the game for itself? Has AlphaGo? I'm confused as to the AI aspect. I understand how it uses MTCS and NN to solve the computation problem, but there's no AI in there, is it?

If you want to simplify how AlphaGo/LZ works, it's kinda like this :

There is the intuitive part of LZ brain : the neural network. LZ see a position, and her intuition (the neural network) give her a list of candidate moves and a feeling of who's ahead.

And there is the reading process : the Monte Carlo search (even if it's not really Monte Carlo anymore, because there is no rollouts). LZ read the most promising moves, and use her intuition to evaluate the position

Her intuition (the neural network) is trained by feeding her millions of self-play games by previous versions of herself, she's told the result of these positions, and her intuition learn what's good (= what's work), and what's bad (what doesn't). And that's how her intuition get better over time.


Now, what's inside the neural network is quite mysterious, but that's not specific to go. It's a "problem" with all deep neural networks. You can train a network to tell if there's a dog in the picture with really high accuracy, but how exactly the neural network recognise the dog is not well understood.

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #122 Posted: Thu Sep 13, 2018 7:40 am 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
Tryss wrote:
And there is the reading process : the Monte Carlo search (even if it's not really Monte Carlo anymore, because there is no rollouts).


OK, now that's confusing: I thought I had to interpret the lower number in Lizzie as "plies", and these plies represent complete rollouts. So now I guess they are not full but partial rollouts, and there's a higher level evaluation than the score.

Tryss wrote:

Her intuition (the neural network) is trained by feeding her millions of self-play games by previous versions of herself, she's told the result of these positions, and her intuition learn what's good (= what's work), and what's bad (what doesn't). And that's how her intuition get better over time.

Now, what's inside the neural network is quite mysterious, but that's not specific to go. It's a "problem" with all deep neural networks. You can train a network to tell if there's a dog in the picture with really high accuracy, but how exactly the neural network recognise the dog is not well understood.


OK. So it's AI after all, not merely an inventive way to speed up reading. Only we get no insight in the "model" used for deciding on either the candidates (explore) or the evaluation of the plie (exploit)

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #123 Posted: Thu Sep 13, 2018 8:01 am 
Dies with sente
User avatar

Posts: 77
Location: Warsaw, PL
Liked others: 165
Was liked: 18
Rank: EGF 3k
Universal go server handle: yakcyll
Knotwilg wrote:
OK. So it's AI after all, not merely an inventive way to speed up reading. Only we get no insight in the "model" used for deciding on either the candidates (explore) or the evaluation of the plie (exploit)

It's not a way to speed up reading, but rather to use prior reading experience to its advantage. Another way, I think a more precise one, to think about the 'intelligence' or the 'intuition' part of a bot is that what it does is not selecting moves based on however 'feeling' could by applied to a program, but rather based on that experience (one could argue that's what intelligence is, but let's avoid that for now). Outside of training, in order to skip the MC search, it employs what's called a value network, which is a neural network used to evaluate positions.

Quote:
Game tree is searched in simulations composed from 4 phases:
  • Selection — simulation traverses tree by selecting edges with maximum action value Q (how good this move is).
  • Expansion — if any node is expanded, it is processed once by SL (Supervised Learning) policy network to get prior probabilities for each legal action.
  • Evaluation — each node is evaluated by value network and by FR (Fast Rollout) policy.
  • Backup — action values Q are updated by values collected during evaluation step.

I recommend this article, it describes how AG works pretty well. Basically, there's no set of rules or knowledge it applies, directly or indirectly; that's our thing. The bot merely collects the data about board positions during learning and formats it so that it can utilize the experience quickly, on the fly - in the form of synaptic weights.


This post by yakcyll was liked by: Waylon
Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #124 Posted: Thu Sep 13, 2018 8:57 am 
Lives in gote

Posts: 337
Liked others: 22
Was liked: 97
moha wrote:
Thanks, this will be interesting. I never saw a statistically significant 40b vs 15b match at more realistic time controls

Here it is :
20 game match between LZ0.15 #157 and #176
--visits=3201 for #176
--visits=12801 for #157
which amounts to approximately time parity (average of 3.03s/move for #176 and 3.4s/move for #157)
no pondering, twogtp V1.4.10, 2x1080Ti
Average game length : 256 moves

#176 wins 13:7
(65% , all games by resignation, 8 wins as W, 5 as B)

Even if 20 games is not enough, it seems you were right for the longer time settings ;-)
Attachment:
157isW.zip [9.82 KiB]
Downloaded 414 times
Attachment:
157isB.zip [9.62 KiB]
Downloaded 406 times


This post by Vargo was liked by 2 people: Gomoto, Waylon
Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #125 Posted: Thu Sep 13, 2018 9:06 am 
Gosei

Posts: 1733
Location: Earth
Liked others: 621
Was liked: 310
Quote:
I recommend this article, it describes how AG works pretty well. Basically, there's no set of rules or knowledge it applies, directly or indirectly; that's our thing. The bot merely collects the data about board positions during learning and formats it so that it can utilize the experience quickly, on the fly - in the form of synaptic weights.


And now please explain how we humans use rules or knowledge to recognize for example an image.

Indeed there is no difference to the bots our brain merely collects the data during learning and formats it so that it can utilize the experience quickly, on the fly - in the form of synaptic ...

It is not that easy to define the difference.

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #126 Posted: Thu Sep 13, 2018 9:08 am 
Gosei

Posts: 1733
Location: Earth
Liked others: 621
Was liked: 310
And while there are no explicit rules in a neural network, we can check the data like sorin and find "rules" the AI adheres to. For example the josekis and moves it prefers in specific configurations.

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #127 Posted: Thu Sep 13, 2018 9:24 am 
Lives in gote

Posts: 311
Liked others: 0
Was liked: 45
Rank: 2d
Vargo wrote:
Even if 20 games is not enough, it seems you were right for the longer time settings ;-)
Thanks, nice to see 40b win at last. This may also answer your earlier question (why official/elo tests are not at "time parity" - no consistent meaning):

Vargo wrote:
40 games between #157 and #176.
Time parity, 5 min per game, GPU: 1x1080, komi 7.5, no pondering.
#157 wins 29:11 (17 wins as W, 12 wins as B)

Vargo wrote:
20 game match between LZ0.15 #157 and #176
--visits=3201 for #176
--visits=12801 for #157
which amounts to approximately time parity (average of 3.03s/move for #176 and 3.4s/move for #157)
#176 wins 13:7 (65% , all games by resignation, 8 wins as W, 5 as B)

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #128 Posted: Sat Sep 15, 2018 3:02 am 
Dies in gote

Posts: 50
Liked others: 0
Was liked: 3
Vargo wrote:
Here it is :
20 game match between LZ0.15 #157 and #176
--visits=3201 for #176
--visits=12801 for #157
which amounts to approximately time parity (average of 3.03s/move for #176 and 3.4s/move for #157)
no pondering, twogtp V1.4.10, 2x1080Ti
Average game length : 256 moves

#176 wins 13:7
(65% , all games by resignation, 8 wins as W, 5 as B)

Even if 20 games is not enough, it seems you were right for the longer time settings ;-)


My test (l0 v15 #157 vs #176, still in progress) :

Code:
C:\APPS\l0gpu\validation.exe -k 157-176 -b C:\APPS\l0gpu\leelaz -n C:\APPS\net\d351f06e.gz -o "-g -v 12801 --gpu 0 --gpu 1 --noponder -t 12 -q -d -r 5 --timemanage off -w" -b C:\APPS\l0gpu\leelaz -n C:\APPS\net\dabff367.gz -o "-g -v 3201 --gpu 0 --gpu 1 --noponder -t 12 -q -d -r 5 --timemanage off -w"

Code:
Stopping engine.
25 wins, 15 losses
40 games played.
Status: 0 LLR 0.821218 Lower Bound -2.94444 Upper Bound 2.94444


P.S. If someone wants the games, I can upload them (after the end of the test).

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #129 Posted: Sat Sep 15, 2018 11:00 am 
Dies with sente

Posts: 108
Location: France
Liked others: 14
Was liked: 18
Rank: FFG 1d
Who won the 25 games?

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #130 Posted: Sat Sep 15, 2018 12:19 pm 
Lives in gote

Posts: 337
Liked others: 22
Was liked: 97
There's a new 256x40b network (#177), a good occasion to see if the result of the last match (157 v 177) still holds.

20 game match between LZ0.15 #157 and #177
--visits=3201 for #177
--visits=12801 for #157
approximately time parity (#157 takes a little more time)
no pondering, twogtp V1.4.10, 2x1080Ti

It's a draw 10:10
(all games by resignation)
So, not as good a result as the last match, but a confirmation that the new networks have caught up with the old 20b (given enough time)
Attachment:
177isW.zip [9.56 KiB]
Downloaded 414 times
Attachment:
177isB.zip [9.32 KiB]
Downloaded 409 times
nbc44 wrote:
My test (l0 v15 #157 vs #176, still in progress) :
Happy to see someone else run matches, thanks ! Looking forward to the final result :)

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #131 Posted: Sat Sep 15, 2018 12:51 pm 
Dies in gote

Posts: 50
Liked others: 0
Was liked: 3
explo wrote:
Who won the 25 games?

#157

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #132 Posted: Sat Sep 15, 2018 2:33 pm 
Lives in gote

Posts: 311
Liked others: 0
Was liked: 45
Rank: 2d
Vargo wrote:
It's a draw 10:10 (all games by resignation)
So, not as good a result as the last match, but a confirmation that the new networks have caught up with the old 20b (given enough time)
Depends on what time is "enough" time. :) (I guess you meant old 15b.) Allowing 6 sec instead of 3 for example, 6400 visits instead of 3200 would likely shift the score some percents in 40b's favor (random variance aside), and so on with even more time.

These scaling effects are the heart of the problem. A more practical question is how much visits would a user get in daily use (on which hardware?) when analysing his games.

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #133 Posted: Sun Sep 16, 2018 1:55 am 
Dies in gote

Posts: 50
Liked others: 0
Was liked: 3
Vargo wrote:
Looking forward to the final result :)


Nothing interesting right now :D :
Code:
68 wins, 46 losses
114 games played.
Status: 0 LLR 1.64871 Lower Bound -2.94444 Upper Bound 2.94444


P.S. I think 12801 visits is too big for this test.

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #134 Posted: Sun Sep 16, 2018 2:25 am 
Lives in gote

Posts: 337
Liked others: 22
Was liked: 97
moha wrote:
I guess you meant old 15b
Yes, 20b networks are for Lc0 , Leela Chess Zero is similar to LZ (description HERE)
The Computer Chess Championship is going on these days HERE, and Lc0 is doing particularly well.

moha wrote:
how much visits would a user get in daily use (on which hardware?) when analysing his games
To get 3200 visits (#177) or 12800 visits (#157), with 2x1080Ti, it's around 3 sec/move. For one dedicated GPU, maybe from 5-6 sec for one 1080Ti to 15-20sec (?)

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #135 Posted: Sun Sep 16, 2018 3:45 am 
Dies with sente

Posts: 108
Location: France
Liked others: 14
Was liked: 18
Rank: FFG 1d
Based on using lizzie, I need around a minute to get 3200 visits on a 40b network. I have a GTX 1050 which I guess is better than what most go players have. Right now most people should rather use #157 if they want to briefly review a game and identify mistakes.

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #136 Posted: Sun Sep 16, 2018 4:29 am 
Lives in gote

Posts: 311
Liked others: 0
Was liked: 45
Rank: 2d
Tests like these could identify the visit points where successive 40b networks overcome the 15b on 4x visits, measuring 40b progress and letting users choose the stronger size for their hardware and time/patience during reviews.

On weak hw this turning point may well remain too high even after substantial 40b training (network strength has less effect during the first bunch of visits, with shallow searches, so the turning point may not decrease too fast in this range), but on 1080ti it seems to be in reach already.

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #137 Posted: Sun Sep 16, 2018 10:58 am 
Lives in gote

Posts: 337
Liked others: 22
Was liked: 97
explo wrote:
Based on using lizzie, I need around a minute to get 3200 visits on a 40b network.
On my laptop (gpu : GTX965M), I've just run such a game with twogtp (#157 at 12801 visits v. #177 at 3201 visits)
Total time was 5702s. for 286 moves, around 20s. per move. I don't know why there's such a difference...
Attachment:
sc.jpg
sc.jpg [ 79.32 KiB | Viewed 9062 times ]

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #138 Posted: Sun Sep 16, 2018 11:33 am 
Dies with sente

Posts: 108
Location: France
Liked others: 14
Was liked: 18
Rank: FFG 1d
Vargo wrote:
explo wrote:
Based on using lizzie, I need around a minute to get 3200 visits on a 40b network.
On my laptop (gpu : GTX965M), I've just run such a game with twogtp (#157 at 12801 visits v. #177 at 3201 visits)
Total time was 5702s. for 286 moves, around 20s. per move. I don't know why there's such a difference...
Attachment:
sc.jpg


I wonder if it has to do with lizzie or the way something is set on my computer. I tried it again, setting an easy opening situation on lizzie with pondering off. Then turning pondering on for a minute. I ended with around 2.8 visits over the moves shown.
Earlier this afternoon, I wanted to try it on GRP to get an idea of how long it would take to review a game with lots of visits on a 40b while I'm at work. I used my last game, asked for 10k visits, and it went much faster than expected. I ended the review prematurely but I think it went through 70 moves in around a hour and a half, which is much faster than what I said about lizzie.
When I installed GRP on my laptop (with an old GPU), it changed something with openCL and things got faster on lizzie later. Is there something special I should install or set up?

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #139 Posted: Sun Sep 16, 2018 9:29 pm 
Lives in gote

Posts: 337
Liked others: 22
Was liked: 97
I've tried with Lizzie V0.5 and network #177, on my GTX965M. It's indeed slow, with times from 35s. to 75s. to get 3200 visits.

Top
 Profile  
 
Offline
 Post subject: Re: LZ's progression
Post #140 Posted: Wed Sep 19, 2018 11:17 pm 
Dies in gote

Posts: 50
Liked others: 0
Was liked: 3
My test (l0 v15, validation.exe .... --gpu 0 --gpu 1 --noponder -t 12 -q -d -r 5 --timemanage off) #157 vs #176 :

I. by visits

a). 12801:3201 (157-176.zip)
Code:
The first net is better than the second
#157 v #176 ( 227 games)
              wins        black       white
#157  134 59.03%   62 59.62%   72 58.54%
#176   93 40.97%   42 40.38%   51 41.46%
                      104 45.81%  123 54.19%
227 games played.
Status: 2 LLR 2.98898 Lower Bound -2.94444 Upper Bound 2.94444

b). 9601:3201 (157-176-1.zip)
Code:
#157 v #176 ( 54 games)
          wins        black       white
#157   31 57.41%   17 56.67%   14 58.33%
#176   23 42.59%   13 43.33%   10 41.67%
                       30 55.56%   24 44.44%
54 games played.
Status: 0 LLR 0.53734 Lower Bound -2.94444 Upper Bound 2.94444

c). 6401:3201 (157-176-2.zip)
Code:
#157 v #176 ( 60 games)
          wins        black       white
#157   22 36.67%   13 38.24%    9 34.62%
#176   38 63.33%   21 61.76%   17 65.38%
                       34 56.67%   26 43.33%
60 games played.
Status: 0 LLR -1.95476 Lower Bound -2.94444 Upper Bound 2.94444

d). 8001:3201 (157-176-3.zip)
Code:
#157 v #176 ( 68 games)
          wins        black       white
#157   35 51.47%   19 51.35%   16 51.61%
#176   33 48.53%   18 48.65%   15 48.39%
                       37 54.41%   31 45.59%
68 games played.
Status: 0 LLR -0.15023 Lower Bound -2.94444 Upper Bound 2.94444


Attachments:
File comment: 6401:3201
157-176-2.zip [56.7 KiB]
Downloaded 373 times
File comment: 9601:3201
157-176-1.zip [52.1 KiB]
Downloaded 419 times
File comment: 12801:3201
157-176.zip [216.48 KiB]
Downloaded 396 times


Last edited by nbc44 on Wed Sep 19, 2018 11:20 pm, edited 1 time in total.

This post by nbc44 was liked by: splee99
Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 418 posts ]  Go to page Previous  1 ... 4, 5, 6, 7, 8, 9, 10 ... 21  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group