Game 3, move 4, Forum 5-3

[Zorba]

I don't like extending the vote period, 2 days is more than enough IMO and everyone can change their vote at every point, so more time won't  really change anything in this process and just slow the match down I'm afraid.

I'm not a fan of the idea that I would decide the best move on subjective criteria, it would mean I would have to value some people's comments higher than others people's. Not a good idea IMO and personally I find it unworkable, as I usually have bot results in front of me by that time. So, instead of tossing a coin when the vote is tied, I think the method I described is at least slightly more satisfying. Encouraging early voting in itself isn't too bad either 

[playBunny]

I dont think that 2 days is more than enough. The tie may not become apparent until late in the time frame and some people may not get to see it and take action. An extra day for further discussion, persuasion and vote changing will slow the match down but it's an assumption that that's a bad thing. I'm certainly not in such a great hurry that I'd rather have an arbitrary decision than one that we've come to ourselves.

Perhaps we 'd benefit from a poll about what tie-breaking method people would prefer?

[ah_clem]

So, instead of tossing a coin when the vote is tied, I think the method I described is at least slightly more satisfying. Encouraging early voting in itself isn't too bad either 

One way to think about your method is to make the result "sticky".  i.e. When the first vote is cast, that choice becomes the presumptive outcome.  As more votes come in, the result doesn't change until another choice gets more votes than it and takes an actual lead - ties do not change the leader.   

I don't know how difficult it would be to implement this, but if you can see timestamps of various votes you can just look at the last timestamp of the tied choices to figure it out.

I'm ok with this method.  Since people can change their vote any time, if it's obvious that your first choice is not garnering much support you can change to your second choice.  No need to wait until voting closes and a runoff period to switch.

[playBunny]

 
Rollout 1296 trials, 2-ply throughout

Spoiler

The voting was well split and GnuBg's rollout reflects that. The top three moves are not well separated by the rollout. The standard deviation of error (Std Err) is large at 0.012 and so Jsds and %Confidence have been given to add perspective to the interpretation.

# .. Move ........... Equity ..(Diff) ... Win ..... WinG .. WinBg -- Lose ... LoseG . LoseBg .. Jsds .. %Confidence
1 .. bar/20 24/21 ... +0.256 ............ 53.1% .. 16.3% ... 1.3% -- 46.9% .. 15.0% ... 2.0%
2 .. bar/20 9/6 ..... +0.247 (-0.009) ... 52.9% .. 16.4% ... 1.1% -- 47.1% .. 15.0% ... 1.6% .. 0.575 .. 72%
3 .. bar/20 13/10 ... +0.235 (-0.020) ... 52.2% .. 17.3% ... 1.2% -- 47.8% .. 15.8% ... 2.6% .. 1.206 .. 88%
4 .. bar/20 6/3 ..... +0.150 (-0.106) ... 50.4% .. 16.0% ... 1.1% -- 49.6% .. 16.6% ... 2.3% .. 6.450 . 100%
Std Err (ave) ........ 0.011 ............. 0.2% ... 0.2% ... 0.1% --- 0.2% ... 0.3% ... 0.2%

For those who are keen, this article, Understanding Uncertainty, will go some way towards explaining what Jsds and % Confdence is about. For those who aren't, the 72% in the above table means that move #1 is probably (72% chance) better than #2 but may not be (28%) and the 88% says that move #1 is more probabably better than move #3 yet still may not (12% chance).

Essentially, the equities that a rollout gives are guesses and the true value is somewhere around. Due to the size of the standard deviation of error, there's an overlap of the ranges covering where the true equities of the three moves may be. This is why the order of the moves is not to be taken as gospel but viewed in terms of that % chance that they are correct.

Further rollout trials would reduce the Std Err and thereby increase the certainty but this was a very slow rollout (and I made a mistake that I only discovered near the end so I had to restart! ) and so I didn't get to extend beyond the 1296.

[close]

[playBunny]

.. the 72% in the above table means that move #1 is probably (72% chance) better than #2 but may not be (28%) and the 88% says that move #1 is more probabably better than move #3 yet still may not (12% chance).

And the 100% means that move #1 is even more probababably better than the #4th move!

[Zorba]

GnuBG 0/2-ply 15552 trials rollout:

Spoiler

I took a different approach, lower play settings but many more trials to get the SE down for statistical significance. A very slow rollout indeed...this took more than 24 hours to complete. Fortunately, both playBunny's and my rollouts are very much in agreement here, so it looks like this is the ordering of the plays indeed. Pretty close call here, it's interesting to take a look at the w/g/bg breakdown; you can see f.i. that the 9/6 move loses least gammons by cleaning up the blot and that 13/10 wins most gammons, being the best priming/attacking move. Leaving two blots also means it wins least games and loses most gammons, which is why it ended up third, despite the matchscore. At a score where losing gammons doesn't matter at all, and winning them is worth twice as much (like 2-away, 1-away Crawford), B/20 13/10 might just be the winner.


    1. Rollout          bar/20 24/21                 Eq.:  +0,2602
        53,02  16,41   1,20 -  46,98  14,69   1,61 CL  +0,1364 CF  +0,2602
      [  0,06   0,06   0,03 -   0,06   0,06   0,06 CL   0,0020 CF   0,0041]
    2. Rollout          bar/20 9/6                   Eq.:  +0,2490 ( -0,0112)
        52,95  16,30   1,06 -  47,05  14,34   1,47 CL  +0,1332 CF  +0,2490
      [  0,06   0,05   0,02 -   0,06   0,07   0,06 CL   0,0018 CF   0,0037]
    3. Rollout          bar/20 13/10                 Eq.:  +0,2427 ( -0,0175)
        52,44  17,25   1,20 -  47,56  15,84   2,26 CL  +0,1243 CF  +0,2427
      [  0,06   0,05   0,03 -   0,06   0,07   0,07 CL   0,0020 CF   0,0040]

Full cubeful rollout with var.redn.
15552 games, Mersenne Twister dice gen. with seed 863347199 and quasi-random dice
Play: 0-ply cubeful prune [expert]
Cube: 2-ply cubeful prune [world class]

[close]

[Zorba]

Thanks for your comments both, on the tied vote 

One way to think about your method is to make the result "sticky".  i.e. When the first vote is cast, that choice becomes the presumptive outcome.  As more votes come in, the result doesn't change until another choice gets more votes than it and takes an actual lead - ties do not change the leader.  

I don't know how difficult it would be to implement this, but if you can see timestamps of various votes you can just look at the last timestamp of the tied choices to figure it out.

Yes, this is the idea. Unfortunately I can't see timestamps on the votes, so it depends on my own observations.

I'm ok with this method.  Since people can change their vote any time, if it's obvious that your first choice is not garnering much support you can change to your second choice.  No need to wait until voting closes and a runoff period to switch.

Yes, this is the idea. So far, it seems that (nearly) all the votes are usually in after 36 hours or so. You can see right after voting what the outcome is, refresh the page later to do so. Then you can always change your mind to break a tie, and vote differently right away or in the remaining time.

I dont think that 2 days is more than enough. The tie may not become apparent until late in the time frame and some people may not get to see it and take action. An extra day for further discussion, persuasion and vote changing will slow the match down but it's an assumption that that's a bad thing.

Yes, it can become a tie at the very end of the voting period, but that's also the reason I don't think extending the voting period will help much; it might again lead to a tie, etc.

Consider this: Suppose you voted play A early, and it leads the poll. Near the end, the vote gets tied. Would you now switch to vote B? Seems unlikely. Now suppose you voted play A early, and play B is leading the poll. Either you switch to play B right away, if you're persuaded, or you don't. Now the vote gets tied at the end. Would you want to switch to play B now? It would win the poll anyway. So, it looks like there's not much to gain by waiting for the tie, and then extending the voting period.

Furthermore, I'm a bit skeptical based on the experience so far, about interesting discussions after the initial two day voting period, I'm afraid the vote will mostly just "sit there" and nothing much will happen. Perhaps if we get more voters and regulars here, this might work, but for now, I think it will just decrease interest.

I'm certainly not in such a great hurry that I'd rather have an arbitrary decision than one that we've come to ourselves.

Perhaps we 'd benefit from a poll about what tie-breaking method people would prefer?

What if that poll gets tied?  

Anyhow, moving the match along is not really hurrying IMO, it's mostly about creating more and hopefully, more interesting decisions to discuss. And I don't think my method is arbitrary, obviously the choice got the most votes from the forum (along with other play(s)) so that in itself is not too bad, but also, if the play has been leading in the vote earlier on, it's a bit more likely that this play is actually the favourite if more voters would come in (not regarding vote changing).

I'm okay with a poll, but I'm not sure what the options should be. Anyone can set a poll up here, so feel free to do so!