Two Plus Two Publishing LLC Two Plus Two Publishing LLC
 

Go Back   Two Plus Two Poker Forums > >

Backgammon Forum hosted by Bill Robertie. Discussion of anything related to backgammon: strategy, problems, books, clubs, and tournaments.

Reply
 
Thread Tools Display Modes
Old 06-15-2021, 01:20 PM   #51
browni3141
Pooh-Bah
 
Join Date: Aug 2015
Location: South Florida
Posts: 4,944
Re: Computers vs Humans

Quote:
Originally Posted by EdwardCollins View Post
I love "Game of the Amazons." (It's also just known as Amazons.)

But no, Amazons wasn't designed to be difficult for computers. I've never read that about Amazons, and if it was, someone did a very poor job. For example, there are only four pieces each and each piece has the same powers. With each move, there are fewer squares on the board that can be moved to. The rules are simple and strong bots already exist.

I strongly suspect you're thinking of the board game Arimaa. According to wikipedia, Arimaa is a two-player strategy board game that was designed to be playable with a standard chess set and difficult for computers while still being easy to learn and fun to play for humans.
The best Arimaa bots are better than humans too. Humans can compete with the best traditional alpha beta bot (bot_sharp) and I'd argue it's not clear who has an edge. Humans lost a match in 2015 to Sharp, and it was convincing enough for me, but development on that bot has ceased since then and IMO humans have gotten a bit better.

However, there is a recent deep learning bot (bot_rusty) which completely trounces every other playing entity so far. Since Rusty has finished training it has only lost a couple games out of hundreds when allowed to play at full strength and they were pretty much flukes that would probably be eliminated with further improvement.

Right now I'm not aware of any strategy games with a significant development effort where humans can compete with computers.
browni3141 is offline   Reply With Quote
Old 06-16-2021, 05:00 AM   #52
bgblitz
newbie
 
Join Date: Mar 2020
Posts: 17
Re: Computers vs Humans

Quote:
Originally Posted by browni3141 View Post
Right now I'm not aware of any strategy games with a significant development effort where humans can compete with computers.
IMHO due to the recent breakthrough, applying image recognition stuff on games (i.e. what DeepMind did with AlphaGO), I feel it is not possible anymore to create a game where bots are far worse than humans. The branching factor as a problem for computers doesn't work anymore with deep neural nets and MCTS.
bgblitz is offline   Reply With Quote
Old 06-16-2021, 05:07 AM   #53
bgblitz
newbie
 
Join Date: Mar 2020
Posts: 17
Re: Computers vs Humans

Quote:
Originally Posted by CTCamp View Post
Thanks again for the answer but again that didn't answer my question. I guess I didn't make myself clear. Here goes the long winded version: Isn't using variance reduction asking the NN to assign weights based upon ITS OPINION of the luck? I am not interested in doing a rollout just to get the same answer that I got from my normal analysis using XGroller++ gigantic more slowly. Speeding up a rollout to get the same wrong answer makes no sense to me. Presumably you are only doing rollouts when you suspect that the analysis returned was wrong. Why else would anyone do a rollout? If anyone has done a rollout using variance reduction and received an answer that differed by more than .01 from an XGroller++ gigantic analysis I would like too know the position so I could experiment for myself.
The answer is simple: VR only speeds up rollouts, it doesn't change the result. If the bot is wrong about a position, he will most often be wrong even after a rollout and the same with VR. The only difference is, if the luck estimation is not totally wrong, you get the results faster.
So yes, VR doesn't help if the bot misjudges.
bgblitz is offline   Reply With Quote
Old 06-19-2021, 11:56 AM   #54
CTCamp
stranger
 
Join Date: Dec 2005
Posts: 14
Re: Computers vs Humans

Thanks. If rollouts return the same answer as analysis why bother?
CTCamp is offline   Reply With Quote
Old 06-20-2021, 09:04 AM   #55
Robertie
old hand
 
Join Date: May 2004
Posts: 1,850
Re: Computers vs Humans

They don't, always. An XGR++ evaluation is very good, but a rollout is somewhat better, particularly in complex positions like back games and very involved middle game positions. In some cases the ranking of the plays can be altered, in other cases the ranking stays the same but the differences between the various plays changes.

Most players rely mainly on XGR++, but will use a rollout when the position is particularly interesting.
Robertie is offline   Reply With Quote
Old 06-20-2021, 01:32 PM   #56
CTCamp
stranger
 
Join Date: Dec 2005
Posts: 14
Re: Computers vs Humans

Thank you for the reply. I still don't understand the advantage of using VR when doing a rollout. I know it is faster but it is still using the same NN to decide the luck factor. I don't understand why the NN is trusted to assess the luck associated with VR when it isn't trusted for the original analysis. I thank everyone who has tried to help me to understand. Your time has value. I guess I give up and will keep running rollouts the slow way when I run them. They are interruptible so I can run them at night with little inconvenience other than my power bill.
CTCamp is offline   Reply With Quote
Old 06-20-2021, 03:58 PM   #57
RolldUpTrips
adept
 
Join Date: Oct 2006
Location: Las Vegas
Posts: 1,176
Re: Computers vs Humans

Quote:
Originally Posted by CTCamp View Post
Thank you for the reply. I still don't understand the advantage of using VR when doing a rollout. I know it is faster but it is still using the same NN to decide the luck factor. I don't understand why the NN is trusted to assess the luck associated with VR when it isn't trusted for the original analysis. I thank everyone who has tried to help me to understand. Your time has value. I guess I give up and will keep running rollouts the slow way when I run them. They are interruptible so I can run them at night with little inconvenience other than my power bill.
I'm not positive, but I THINK that VR simply means that the rolls will occur as often as they're "supposed" to; so, if you did a 36-game rollout hypothetically speaking, the first roll in the position would be 11 once, 12 twice, etc. Again, I'm not sure about this but this is how I've interpreted it.
RolldUpTrips is offline   Reply With Quote
Old 06-20-2021, 04:50 PM   #58
_Z_
grinder
 
_Z_'s Avatar
 
Join Date: Jul 2004
Posts: 619
Re: Computers vs Humans

Variance reduction is a way to use evaluations to "speed up" rollouts. That is, if a regular rollout takes 10,000 games to converge to a particular result, a rollout with variance reduction will converge much faster to the same result, say after 1,000 games.

If the evaluations are flawed, it doesn't affect what result the rollout will converge to, it just means it will take more games to converge.

Here's an analogy I thought up to convince myself it works. bgblitz can comment if the analogy is flawed in some ways, but I think it gets across the main point.

Quote:
Let's say you want to figure out what the average roll of one die is, but you don't know how to do the calculation. You could do a rollout -- roll a die a 100 times and you'd get the following rolls in roughly equal proportion:

1 2 3 4 5 6

Then you would average those 100 rolls and it would be around 3.5.

But let's say you had an evaluation function that knew three things --

6 is a little higher than average
1 is a little lower than average
6 occurs exactly as often as 1

Then you could do the same experiment again, but every time you see a 6, you know it's a little too high, so you subtract 1 from it. And when you see a 1, you know it's a little low, so you add 1 to it. This won't affect the average because you have a 1 in 6 chance of adding 1 and a 1 in 6 chance of subtracting 1. So now you get the following rolls in roughly equal proportion:

2 2 3 4 5 5

They still average out to 3.5, but they will converge faster to 3.5 than in the original experiment, as they're more closely clustered around 3.5. In other words, you'll need fewer rolls to get an accurate result.

Now what happens if your evaluation function is bad? Let's say it dumbly thinks 6 is too low and 1 is too high. Then after adjusting, you get the following rolls in roughly equal proportion:

0 2 3 4 5 7

They still average out to 3.5, but it will take longer to converge. An inaccurate evaluation function won't affect the final result, it will just affect how long it takes to converge on it. Luckily, XG's evaluation function is pretty good.
_Z_ is offline   Reply With Quote
Old 06-22-2021, 01:20 PM   #59
bgblitz
newbie
 
Join Date: Mar 2020
Posts: 17
Re: Computers vs Humans

Quote:
Originally Posted by CTCamp View Post
Thank you for the reply. I still don't understand the advantage of using VR when doing a rollout. I know it is faster but it is still using the same NN to decide the luck factor.
Hm, if I need 2 hours instead of 24 to get the same result is no advantage? Even if you have plenty of time you will lower your electricity bill....
bgblitz is offline   Reply With Quote
Old 06-22-2021, 01:24 PM   #60
bgblitz
newbie
 
Join Date: Mar 2020
Posts: 17
Re: Computers vs Humans

Quote:
Originally Posted by RolldUpTrips View Post
I'm not positive, but I THINK that VR simply means that the rolls will occur as often as they're "supposed" to; so, if you did a 36-game rollout hypothetically speaking, the first roll in the position would be 11 once, 12 twice, etc. Again, I'm not sure about this but this is how I've interpreted it.
You are correct in by controlling the dice you reduce variation, but you can't do this for a whole rollout (AFAIK only controlling the first 2 rolls is done). A short explanation is here: https://www.bkgm.com/gloss/lookup.cg...ance+reduction
bgblitz is offline   Reply With Quote

Reply
      

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off


Forum Jump


All times are GMT -4. The time now is 10:41 PM.


Powered by vBulletin®
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.
Copyright 2008-2020, Two Plus Two Interactive