It is currently Thu Mar 28, 2024 8:43 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 66 posts ]  Go to page Previous  1, 2, 3, 4  Next
Author Message
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #41 Posted: Thu Feb 08, 2018 3:40 am 
Lives in gote

Posts: 340
Location: Spain
Liked others: 181
Was liked: 41
Rank: Low
gennan wrote:
Do you think that DeepMind cherry-picked mostly white wins, but in fact unpublished wins had an even color distibution?

DeepMind's Julian Schrittwieser said:

"In my experience and the experiments we've run, komi 7.5 is very balanced, we only observe a slightly higher winrate for white (55%)."

https://www.reddit.com/r/MachineLearnin ... r/doljugm/

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #42 Posted: Thu Feb 08, 2018 11:59 am 
Lives in gote

Posts: 476
Location: Netherlands
Liked others: 270
Was liked: 147
Rank: EGF 3d
Universal go server handle: gennan
luigi wrote:
gennan wrote:
Do you think that DeepMind cherry-picked mostly white wins, but in fact unpublished wins had an even color distibution?

DeepMind's Julian Schrittwieser said:

"In my experience and the experiments we've run, komi 7.5 is very balanced, we only observe a slightly higher winrate for white (55%)."

https://www.reddit.com/r/MachineLearnin ... r/doljugm/


Thanks, I did not know that statement. 45% winrate for black is much closer to 50% than 24%. Still, 45% does favor white enough to suggest a komi error of about 1.5 points if it were statistics from human pro games.

I tried to find more details about that statement, but besides some quotes, I couldn't find any. For example, did DeepMind use the same time settings for the published and unpublished games?

DeepMind used 10 minutes time per position to generate the learning tool winrates, which is much longer than the time limits used in self-play games AFAIK. I can imagine that with 10 minutes per move in self-play games, black would win less than 45% (tending to follow the winrate trends in the learning tool more closely), because the moves would be closer to perfect and thus black would get less chances to upset white's lead, making the effect of oversized komi more pronounced.

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #43 Posted: Thu Feb 08, 2018 4:49 pm 
Lives in gote
User avatar

Posts: 392
Liked others: 23
Was liked: 43
Rank: NR
gennan wrote:
But with an integer komi, there can be a komi value where perfect players would always get a jigo.
That's not logical - the only thing that's sure is that there can be a komi value where imperfect players would sometimes get a jigo. i would imagine that DM experimented with 6.5 and 7.5 and found that 7.5 was closer to 50%. It's entirely possible that 7.5 is closer to 50% than 7.

But one thing is for sure: if anyone ever learns anything from A0, it won't be anything to do with komi.

To me, the most fascinating thing is the markedly different styles of A0 and Master - but of course, i am biased like hell, because A0's honte style is more like Swim's than Master's :D

_________________
i shrink, therefore i swarm

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #44 Posted: Fri Feb 09, 2018 3:23 am 
Lives in gote

Posts: 476
Location: Netherlands
Liked others: 270
Was liked: 147
Rank: EGF 3d
Universal go server handle: gennan
djhbrown wrote:
i would imagine that DM experimented with 6.5 and 7.5 and found that 7.5 was closer to 50%.

To do that DeepMind would have had to raise different instances of AlphaGo for different komi values. I think they would have mentioned it if they did that and reported their findings. But I don't think they did, because of the costs involved in multiplying the expensive hardware costs.

djhbrown wrote:
But one thing is for sure: if anyone ever learns anything from A0, it won't be anything to do with komi.

I think the komi value is quite important at a high level of play. The move choice in a situation depends not only on possibilities on the board, it also depends on the who is leading. When behind (even by a little bit), strong players try to reverse the game with risky moves. When ahead, strong players try to consolidate their lead with safe moves.

The stronger the players, the more accurate their awareness of who is leading in a particular situation and in close games (as games between evenly matched strong players tend to be) komi size is an important factor in this evaluation. Even the first moves in a game between strong players are determined by komi size. With a large komi, black tends to play a speedy opening and white tends to play a steady opening. The reason is that strong players feel that black cannot afford a steady opening if the komi is as large as 7 points. But in the classical era before komi, black played a steady opening and white played a speedy opening. So strong players feel komi size throughout the game and it affects their moves right from the start.

So I would think that komi size is also quite relevant for the moves that AlphaGo recommends in a particular situation. If you watch Michael Redmond's reviews, you'll find a returning theme in AlphaGo vs AlpahGo games that black AlphaGo shows signs of desparation already in the middle game. Black AlphaGo tends to turn a close game into a complicated fight and eventually collapses against white AlphaGo. The cause for black's choice seems to be that black AlphaGo realizes that he will just fall short to pay the full komi. If the komi were one point less (so black only needs 7 points more on the board to win instead of 9, with Chinese rules), it is quite possible that black AlphaGo would choose a safer continuation.

If AlphaGo were trained for no komi games, I think it too would play quite differently. So komi may seem irrelevant for what we can learn from AlphaGo, but indirectly it affects what AlphaGo tells us to do.


This post by gennan was liked by: Uberdude
Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #45 Posted: Fri Feb 09, 2018 4:28 am 
Lives in gote
User avatar

Posts: 392
Liked others: 23
Was liked: 43
Rank: NR
gennan wrote:
Black AlphaGo tends to turn a close game into a complicated fight and eventually collapses against white AlphaGo.... it affects what AlphaGo tells us to do.
i don't want to be a pedantist, but please can you clarify which version of Alfie you refer to? My guess is it's Master v Master, which i regard as being two red herrings squabbling over random rollouts.

i'm no judge, but it does look to me like Alfie0 is calm and collected, whichever colour she takes.

_________________
i shrink, therefore i swarm

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #46 Posted: Fri Feb 09, 2018 5:16 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
djhbrown wrote:
i'm no judge, but it does look to me like Alfie0 is calm and collected, whichever colour she takes.

djhbrown wrote:
... because A0's honte style is more like Swim's than Master's :D

I'm surprised by your characterisation of AG0 as calm and honte: I'd say it's more greedy, aggressive in taking territory, and sharp, playing moves (particularly invasions) that I'd have thought were overplays but somehow its deep reading means it can get away with it.


This post by Uberdude was liked by: gennan
Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #47 Posted: Fri Feb 09, 2018 6:33 am 
Lives in gote
User avatar

Posts: 392
Liked others: 23
Was liked: 43
Rank: NR
Uberdude wrote:
I'd say it's more greedy... invasions that I'd have thought were overplays
i've only looked at Redmond's Reviews, and i recall that A0 made an invasion in game 4 that took even him by surprise, but as far as i remember he pointed out that it had two escape routes, so i got the feeling that A0 had played calmly, biding her time until just the right moment.

Also, i remember people talking about a game where A0 made Master look like a donkey, neatly living inside a heavily walled yet overstretched moyo.

And i will never forget the sabaki elegance A0 demonstrated on the left side in Redmond's Reviews game 1, which impressed me hugely. Of course, i've forgotten all the details, but i remember the emotion i felt on seeing it.

As to the term "greedy", that strikes me as signifying 'going after more than you can swallow', ie an overplay - and by definition, overplays don't work. I'm not aware of A0 making any overplays, but it is conceivable, given that {kneejerk reactions + PHTS} theoretically cannot be always perfect.

_________________
i shrink, therefore i swarm

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #48 Posted: Fri Feb 09, 2018 4:02 pm 
Lives in gote

Posts: 476
Location: Netherlands
Liked others: 270
Was liked: 147
Rank: EGF 3d
Universal go server handle: gennan
I would call A0 patient rather than calm. And it can afford its patience, because it's so accurate in its evaluation / judgement. And that is based on its extremely accurate reading.

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #49 Posted: Fri Feb 09, 2018 4:17 pm 
Lives in gote
User avatar

Posts: 392
Liked others: 23
Was liked: 43
Rank: NR
gennan wrote:
it can afford its patience
sorry, i can't resist this one :) Do you realise what you just said?!

"afford patience"

that says more about you than it does about Her Alfness :mrgreen:

i can afford to be lazy now, because i worked my knuckles to the bone for 52 years. But patience is a virtue, and virtue means (ie used to mean) "manly, warriorlike"

Quote:
c. 1200, vertu, "moral life and conduct; a particular moral excellence," from Anglo-French and Old French vertu "force, strength, vigor; moral strength; qualities, abilities" (10c. in Old French), from Latin virtutem (nominative virtus) "moral strength, high character, goodness; manliness; valor, bravery, courage (in war); excellence, worth," from vir "man" (from PIE root *wi-ro- "man").

For my part I honour with the name of virtue the habit of acting in a way troublesome to oneself and useful to others. [Stendhal "de l'Amour," 1822]

_________________
i shrink, therefore i swarm

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #50 Posted: Fri Feb 09, 2018 4:48 pm 
Lives in gote

Posts: 476
Location: Netherlands
Liked others: 270
Was liked: 147
Rank: EGF 3d
Universal go server handle: gennan
djhbrown wrote:
"afford patience"

that says more about you than it does about Her Alfness :mrgreen:

i can afford to be lazy now, because i worked my knuckles to the bone for 52 years. But patience is a virtue, and virtue means (ie used to mean) "manly, warriorlike"


I don't quite understand why it says more about me than A0, but I think there is truth in what you say afterwards. A0 plays lazily, because it knows that it can. But even A0 is wrong sometimes, because Master did win some games.

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #51 Posted: Fri Feb 09, 2018 6:33 pm 
Lives in gote
User avatar

Posts: 392
Liked others: 23
Was liked: 43
Rank: NR
gennan wrote:
why it says more about me than A0
your question opens up a can of worms so deep it makes Alfie's iceberg look like a pimple.

but it's such an important question - so important that schoolteachers of English and History and Politics and and and should maybe spend more time on it than telling us how to solve quadratic equations, or funny things about Pi.

Like all natural languages, English is so elliptical that anything anyone says could mean just about anything, and when reading or listening what one sees or hears is more a function of oneself than the other party, let alone what they say, which invariably isn't what they mean anyway...

Which all goes to prove that Watson was a complete waste of time and could end up doing real damage if they let it loose in doctor's offices -

- Because, to make sense of anything, you have to infer what it could mean, just as Alf has to infer, not what opp means, but how to kill him/her regardless of what they say next.

So, to me, the word afford means a luxury - something you don't need. I can't afford not to breathe, but i can afford to smoke (except i can't, because it's killing me, but that's another story).

i read the phrase: "[Alfie] can afford patience" as suggestive that the speaker (in this case, you) imagines that patience is a luxury; that patient play is luxuriously lazy, as your followup suggests.

that suggests to me that when you play Go yourself, you feel you don't have the luxury of patience, because you are flat out firefighting all the slings and arrows of outrageous fortune that are flying at you from the other side of the board - or, perhaps, too busy flinging them yourself.

Or both.

I just finished a game in which my (as i fondly like to think of it) ordinary patience was briefly set aside, when a bully descent opportunity turned up and i felt frivolous enough to turn from Jekyll to Hyde just for fun.
Attachment:
11614193-307-dw197-myopian.sgf [2.5 KiB]
Downloaded 449 times

So now there were two Hydes in the ring, both flailing about like whirling dervishes.



i quickly returned to my normal self, but Hyde White didn't change his spots, and it went on, and on, and on.

my confidence surged with each nonsense and my grin became broader than the Cheshire Cat's.

Until the end, when - and i hadn't even known it, not until they told me - that the score was
Hyde 1, Jekyll less than nothing.

Jesus wept, i never will understand this game.

But back to you, and patience.

Whereas all the books i read back in the Dark Ages extolled the virtues of patience in the Protracted Game, which more than anything caught my imagination and my fancy - to me, it is the whole appeal of Go, that here is a game in which patience is clearly not just an asset, but a vital platform on which to stand.

Unfortunately, whereas i can relish and applaud the patience of Alfie0 - and laugh at the impatience of Alfie Master - although i have patience when playing, it doesn't get me anywhere; maybe i have too much of it, neglecting the present because i only have eyes for the future, not seeing that i'm falling off a cliff until i actually hit the ground.

Just a minute! this is supposed to be about you, not about me. Sorry.

Enough about me - let's talk about you; what do you think of me? :D

So,..., if you feel patience is something only the rich can afford, maybe that implies you're missing the bigger picture.

Or i am - but like i said, this is not about me, it's about you.

PS. in this next game, black bided his time patiently, waiting for an opportunity, whilst white tried one trick after another. Although black's opportunity never came (white was too canny for that), black did much better than in the previous one, even though still lost. Any suggestions? (perhaps in another thread, so as to not drag this one off-topic).
Attachment:
11615187-330-chncloud-myopian.sgf [2.58 KiB]
Downloaded 454 times

Go keeps me off the streets, but my hedonistic patience at it doesn't make me virtuous (Eudaiamonia or however you spell it [a rough diamond?]):




Gentle breath of yours my sails
Must fill, or else my project fails,
Which was to please.
Now I want
Spirits to enforce, art to enchant,
And my ending is despair,



_________________
i shrink, therefore i swarm

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #52 Posted: Sat Feb 10, 2018 6:24 pm 
Lives in gote
User avatar

Posts: 603
Liked others: 43
Was liked: 139
Rank: 6-7k KGS
djhbrown wrote:
gennan wrote:
But with an integer komi, there can be a komi value where perfect players would always get a jigo.
That's not logical - the only thing that's sure is that there can be a komi value where imperfect players would sometimes get a jigo. i would imagine that DM experimented with 6.5 and 7.5 and found that 7.5 was closer to 50%. It's entirely possible that 7.5 is closer to 50% than 7.

But one thing is for sure: if anyone ever learns anything from A0, it won't be anything to do with komi.

To me, the most fascinating thing is the markedly different styles of A0 and Master - but of course, i am biased like hell, because A0's honte style is more like Swim's than Master's :D

Completely disagree. Proof:

1) There is such a thing as perfect play. After all, one (well, more due to symmetry) move must be the best in any given situation (or multiple moves could be equal in that they lead to the same outcome in terms of score)
2) Both sides play every perfect move
3) The game will end with each side having some integer number of points.
Therefore,
4) There must then be some integer komi that would make the final score equal.


This post by Fedya was liked by: gennan
Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #53 Posted: Sat Feb 10, 2018 7:41 pm 
Lives in gote
User avatar

Posts: 392
Liked others: 23
Was liked: 43
Rank: NR
Fedya wrote:
Proof
there's a flaw in your straw, dear Liza, dear Liza; there's a flaw in your straw, dear Liza, A Flaw!

Hint: It's got something to do with the Monty Hall Problem

_________________
i shrink, therefore i swarm

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #54 Posted: Sun Feb 11, 2018 6:26 am 
Lives in gote
User avatar

Posts: 603
Liked others: 43
Was liked: 139
Rank: 6-7k KGS
Why don't you actually post the flaw instead of a constant stream of useless non sequitur garbage?


This post by Fedya was liked by 4 people: bernds, Joaz Banbeck, luigi, Waylon
Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #55 Posted: Sun Feb 11, 2018 7:58 am 
Lives in gote
User avatar

Posts: 392
Liked others: 23
Was liked: 43
Rank: NR
Your avatar is Charles Coburn.

Wikipedia wrote:
In the 1940s, Coburn served as vice-president of the Motion Picture Alliance for the Preservation of American Ideals, a group opposed to leftist infiltration and proselytization in Hollywood during the Cold War.[citation needed] Coburn was a member of the White Citizens' Councils, a group which opposed racial integration.[4][5]


in my eyes, black and white are equal.

_________________
i shrink, therefore i swarm

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #56 Posted: Sun Feb 11, 2018 8:11 am 
Lives in gote

Posts: 476
Location: Netherlands
Liked others: 270
Was liked: 147
Rank: EGF 3d
Universal go server handle: gennan
djhbrown wrote:
It's got something to do with the Monty Hall Problem

I would be interested in a clear explanation why perfect players would not always get a jigo if the komi is perfect. I don't really see why the Monty Hall Problem disproves it. Can you clarify your hints? Do you mean that the perfect komi cannot exist?


Last edited by gennan on Sun Feb 11, 2018 8:49 am, edited 4 times in total.
Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #57 Posted: Sun Feb 11, 2018 8:42 am 
Lives in gote

Posts: 476
Location: Netherlands
Liked others: 270
Was liked: 147
Rank: EGF 3d
Universal go server handle: gennan
Fedya wrote:
Why don't you actually post the flaw instead of a constant stream of useless non sequitur garbage?

I'm not interested in trolling either (yes, I too consider the rest of post #53 to be trolling), so I can understand the angry response of post #54. But I think the best reaction to trolling is just ignoring it. At least, that's what I try to do when discussing on internet fora.

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #58 Posted: Fri May 04, 2018 1:25 am 
Lives in gote

Posts: 476
Location: Netherlands
Liked others: 270
Was liked: 147
Rank: EGF 3d
Universal go server handle: gennan
I made a new episode: Learning from AlphaGo #6: the Kobayashi opening


This post by gennan was liked by: daal
Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #59 Posted: Sun Sep 30, 2018 3:05 pm 
Lives in gote

Posts: 476
Location: Netherlands
Liked others: 270
Was liked: 147
Rank: EGF 3d
Universal go server handle: gennan
I made a new episode: Learning from AlphaGo #7: Some AlphaGo guidance in a kyu level opening.

Top
 Profile  
 
Offline
 Post subject: Re: Video series: Learning from AlphaGo
Post #60 Posted: Mon Oct 08, 2018 4:08 pm 
Lives in gote

Posts: 476
Location: Netherlands
Liked others: 270
Was liked: 147
Rank: EGF 3d
Universal go server handle: gennan
I just published a new video:

Learning from AlphaGo #8: Zero style, kick and invade!

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 66 posts ]  Go to page Previous  1, 2, 3, 4  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