Re: [Bug-gnubg] Fwd: Listing JSDs in rollouts

[Ian Shaw]

Philippe wrote, 

"Some details are missing, as long as one can switch back to the detailed 
output, it doesn't really matter"

The missing details are a problem once you've exported the data, and perhaps 
posted the rollout on a website. If you are still within gnubg you can check 
the settings by getting a new export*. Once you've posted the brief data to a 
website then other users can't check the settings.  

*It's not even intuitive how to review the settings for a rollout even within 
gnubg. How have to know to use the Copy option.

I agree that the footnote technique is not perfect, but I do think it's much 
better than what we've got now. Having looked any many rollouts at 
bgonline.org, I find XG's style much easier to read than gnubg's or Snowie's. 
It presents the data in an order much more suited to the thought process I am 
interested in. 

There is always a tradeoff being made. Gnubg keeps all the information about a 
rollout together, at the cost of separating the plays being compared. XG keeps 
the rollout results together, at the cost of separating them from the rollout 
settings. 

I find XG's approach the best so far.

For example, if I've rolled out initial 21: 13/11 6/5 and 13/11 24/23 I want to 
look at the following, in approximate order:

(a) the equity of the two plays (and the difference)
(b) the comparison of wins, losses and gammons, where slotting gains and loses 
over splitting
(c) the JSD in the rollout, to see if the difference is significant
(c) the rollout settings

I've found this much easier with an XG rollout, which also lists the w/g/b 
breakdown on 2 lines, something like:

1       R       8/5 6/5         -0.6879
Player:   0.3462        G:0.0792        B:0.0030        
Opponent: 0.6538        G:0.2416        B:0.0096
5185 games, play 2-ply, cube 2-ply (1)

2       R       24/23 8/5       -0.7443 ( -0.0564  2.345 JSD)
Player:   0.3164        G:0.0637        B:0.0024  
Opponent: 0.6836        G:0.1566        B:0.0081
5185 games, play 2-ply, cube 2-ply (1)

I do like your idea of including the JSD next to the equity difference.

I also like the idea of including the ply information along with the games; 
it's not really taking up any extra space.

So I suggest:
Include the trials and ply with each play, then footnote the detailed rollout 
settings, as shown above.

I would not bother having an option between old and new style, because I think 
the new style is a clear improvement. It might be worth having the option of 
including the full rollout settings as a footnote, or only report the terse 
trials and ply.

On to other details:

(I've omitted the individual SD data from the suggestion above. If we want to 
keep it, I'd have it to the right:
1       R       8/5 6/5         -0.6879
Player:   0.3462        G:0.0792        B:0.0030        SD: 0.0010, 0.0004, 
0.006
Opponent: 0.6538        G:0.2416        B:0.0096        SD: 0.0010, 0.0004, 
0.006
5185 games, play 2-ply, cube 2-ply (1)

XG omits SD and uses the right-hand side to report the cubeful CI and the 
duration:     
Conf.: ± 0.002 (+0.106...+0.110) - [99.7%]
Duration: 1 day 00 hour 35 minutes

I'm not greatly concerned about reporting SD or CI. It's a minor issue compared 
to the overall layout. The CI ± 0.002 is quite useful. Reporting the 
(+0.106...+0.110) is less useful - I hope the interested backgammon user can do 
that in their head. The [99.7%] is, I think, the probability that this play is 
the "best", which I think is what the user really wants to know. (More 
accurately, it's the probability of this play coming top of an infinitely long 
rollout.)
SD is more flexible than CI, because you can use it to calculate any CI you 
wish. However, most people are interested in 95% CI, so maybe it makes sense to 
calculate it for the user. 

Reporting the duration is more about marketing XG's speed advantage than about 
the rollout. 

XG's use of background colour to separate close plays, errors and blunders is 
also good.

-- Ian

-----Original Message-----
From: Philippe Michel [mailto:address@hidden 
Sent: 31 May 2011 21:29
To: Ian Shaw
Cc: address@hidden
Subject: Re: [Bug-gnubg] Fwd: Listing JSDs in rollouts

On Tue, 31 May 2011, Ian Shaw wrote:

> Gnubg has a larger problem in it's rollout reporting.

> The original is on the left. Look how messy it is.
>
> On the right I have tweaked it to reduce redundant information, and 
> include the suggested JSD information. I think the primary data - the 
> rollout results - are a lot more legible.
>
> What do people think of this?

I agree on the general idea, but don't like the "settings as footnote" 
trick at all.

I'd rather have a setting to choose between the original output and 
something more terse than your example (whithout the sd line for 
instance). Something like :

1       R       8/5 6/5         -0.6879
0.3462  0.0792  0.0030  -       0.6538  0.2416  0.0096  -0.4766 -0.6879
5185 games, play 2-ply, cube 2-ply

2       R       24/23 8/5       -0.7443 ( -0.0564  2.345 JSD)
0.3164  0.0637  0.0024  -       0.6836  0.1566  0.0081  -0.4660 -0.7443
5184 games, play 2-ply, cube 2-ply

Some details are missing, but as long as one can switch back to the 
detailed output, it doesn't really matter.