Skip to content
Little GolemMake your next move.

Super Tic Tac Toe

Public forum discussion.

General

#52 · FatPhil

12 Feb 2014

Ingo: your http://www.3-hirn-verlag.de/dreistein/dri-42.png was a snapshot of my bug (I'd rewound my move history so I could test win detection again, but forgot to clear the X flag from the summary). Can you delete it from the record please, the image after it shows what should have been seen.

Handy perl hint - this doesn't count the number of O's or X's in the $outer string
$c=$player?'X':'O';
if(($outer =~ tr/$c/$c/))>=2 { $score=$player?"0:3":"3:0"; }
:-(
(It counts the number of $'s and c's instead)

#53 · FatPhil

12 Feb 2014

Funny - textile has turned what I typed (which was 2 strings:
quote zero colon three quote
quote three colon zero quote
Into something which not only parses, but almost looks sensible!

Don't be tempted to follow the link it created ;-)

#55 · FatPhil

13 Feb 2014

The inner EWN can't be evaluated the same way as normal, as you don't have to play in a board. E.g. If you can engineer ending up with one piece each, then if your piece is in the corner, you win by never moving until the enemy piece moves next to you. It messes up with the zugzwang.

I've realised that the simplest pure race-oriented variant is 2 boards, first to 1. Should I implement that too? (I probably will anyway, as it's to simple. I guess it will take 6 minutesn, but it's late.)

#56 · Ingo Althofer

13 Feb 2014

Hi Phil,

> I’ve realised that the simplest pure race-oriented variant is 2 boards,
> first to 1.

That went also through my head this night.

> Should I implement that too?

Please, do so.

And I would be volunteer for the first testgame (against you or
anyone else).

Thanks for your implementations!

Ingo.

#57 · FatPhil

13 Feb 2014

Just woke up. 12 minutes to code, 6 minutes to debug. Yay!

We're game 29.

Anyone can start their own games at will (I just hope the web-crawlers don't start getting too over-excited, or I'll need a dumb capture type thing). Is luiser still a quick-play EWN monster? Perhaps I should invite her to a game or 20...

Of course, there's no tracking of identities, no ratings, no reason to play except for the process of discovery and the fun which that entails, which is for me the most important thing.

As the selection of boards within which the inner EWNs can be played has exploded, I've renamed the family of metagames "AllBoard EinStein". :-D

#58 · Carroll

13 Feb 2014

Let call luck factor a number between -1 and 1: lf(G) = outcome(G)-p(win) where outcome is 1 if player wins, 0 otherwise. It has the nice property that lf(G) for player B is minus lf(G) for player A.

This is valid for any state of the game if you can compute the probability to win.
For games where each move is as important as an other (no tie break), I guess sum lf(Gi) is a significant luck measure, where you take for outcome the new probability to win.

You travel on a luck factor path to get in the end (outcome-p0) !

#59 · Carroll

13 Feb 2014

FatPhil unlucky messages Oooooh and Ooooh, represent luck factors of -5/6 and -4/6.
You can't complain on luck factors between -1/2 and 1/2 which are pretty standard.

#60 · Ingo Althofer

13 Feb 2014

Carroll, good one.

> FatPhil unlucky messages Oooooh and Ooooh, represent luck factors
> of -5/6 and -4/6.
> You can’t complain on luck factors between -1/2 and 1/2 which
> are pretty standard.

Perhaps, Richard M should introduce some code of conduct for comments
in games with chance. Making inappropriate comments for situations
with LF near to zero leads to a yellow card. In case of repeated
mis-conduct the number opf vacation days might be reduced.

Ingo (still laughing).

#61 · FatPhil

13 Feb 2014

p(win) varies all during the game
IT's "ooooops", not "oooooh"

#62 · FatPhil

13 Feb 2014

Deleted over 100 lines of code in the last couple of hours. That's a very productive half day, IMHO.

More simplification to some, which should make the code even more modular and expandable. (Yes, I foresee more variants...)

#63 · Carroll

13 Feb 2014

@FatPhil yes ooooops!

Yes pi changes so you get :

(p1-p0) + (p2-p1) + ... + (outcome-pn-1) = outcome-p0

the key lucky or unlucky moments are when you go from near to 0 to near to 1 probabilities.

#64 · FatPhil

13 Feb 2014

P0 = 50%

So at the moment this isn't a very good metric.

#65 · Carroll

13 Feb 2014

I don't understand what you mean with a good metric.

As you say you start with around 0.5 (depending if you are a good player, you could be a bit above that). In the end either you lose and get 0-0.5 and you were half unlucky, or you win and get 1-0.5 and were half lucky.

During the journey you can swith from one of these values to the other with some luck moves from you or your opponent.

#66 · FatPhil

13 Feb 2014

I think if I play optimally (given a priori probabilities) at every move, end up with a 35/36 winning position, and then lose, then my luck rating should not be -0.5 (or -1).

Your argument might be that I must have had some good luck earlier in order to get into the 35/36 position, but what if they opponent is not very good, and I played to his weaknesses?

#67 · Carroll

13 Feb 2014

There has been a lot of discussions on luck and skill for poker, and can you tell them apart.

The view I like on this problem is that the two metrics are orthogonal, so there is both luck and skill but you can't say there is more one than the other.
Usually, to win a tournament with many contestants, you need both skill and luck.

Maybe what we would need is two measures to separate your luck and your opponent's luck. For example your probability to win could go to 1 with luck, and then back near 0 with your opponent big luck, and then back to 1, many times, without you being especially unlucky.

#70 · MarleysGhost

13 Feb 2014

> So at the moment this isn't a very good metric.

I think the p[x] has not been precisely defined. I would like to see p[x] as the probability that some standard non-deterministic model player has of winning the game against itself as an opponent given what the game's die rolls have been for the first x moves. Inter alia, it's independent of the current state of the board. That makes outcome-p[last] a measure of player skill, and p[x]-p[x-1] a measure of how good a die roll is.

I fear that others may have interpreted p[x] as the probability that such a player has of winning given the state of the board after x moves, which reflects both the die rolls and the accumulated deviations from average of the moves selected by the players.

#71 · William Fraser

14 Feb 2014

What a cool discussion this is.

I have had some thoughts, but no real time to chase them. My initial reaction was to suspect that MarleysGhost's interpretation will not turn out to be the correct one, although I see arguments on both sides.

Has anyone looked to see what the state of research is in this area is? (Note that, unlike poker, there is no hidden information/blufffing) I'd try drawing an analogy to Yahtzee, which I vaguely seem to recall seeing any article on....

#72 · Carroll

14 Feb 2014

Yes we need a bibliography, some people must have thought seriously on this problem, I skipped through the aforementioned thesis, but think he makes things over complicated.

For me, the probability pi is the true probability to win for player A at stage i. It can be estimated by a program. Then B will throw the dice and play, and then p(i+1) is the new probability indicating A's luck factor (by difference with pi) if both players played perfectly.

#73 · Carroll

14 Feb 2014

There is an article by Douglas Zare on luck in backgammon. He takes the difference in equity as a measure of the luck for one move. This amounts to the same thing as difference of probability taking the value of a win to be one.
I think you could measure skill factor by the total amount of equity you left away on all the suboptimal moves you have made. If you went from a LF of 0.5 to zero, with a bad true LF of 0.3, then the remaining 0.2 must be bad skill factor (SF). SK is always negative, but I don't know if there is a limit on its value.

#74 · Carroll

14 Feb 2014

In fact I defined a negative skill factor or Dumb Factor ...

#75 · FatPhil

14 Feb 2014

I Think there are at least two measures. How much you threw away by playing sub-optimally is certainly one that I've thought of trying to measure in EWN. That's basically what the Dumb Factor seems to be.

However, it doesn't take into account the dice - the luck aspect. It could be that the best metric evaluating luck is just the difference between outcome and dumbness.

But not necessarily. There is a difference between an endgame where it's mostly p(win)=near 0.5 each move (I sometimes mention coint-tosses towards the end of frames, you may have noticed), and a game that fluctuates wildly between p(win)=near 1 and p(win)=near 0 (my *phew!*s and oooopses). i.e. it's not the destination that determines the luck, it's how you got there. But formulating that is a challenge.

#76 · Carroll

14 Feb 2014

Looking again on luck measurement for backgammon best programs, I found a nice article I put in extenso:

Here is how Snowie calculates the luck factor, (and I'm pretty sure GnuBG
does it the same way):

It takes the equity of the 36 positions reached from the current position,
(based on moving each of the 36 rolls "correctly"), and averages those 36
numbers. It then takes the equity reached from the roll actually rolled,
and the amount this is above or below the average is how lucky or unlucky
that roll was. It does this for every roll of the game for both players
(all of this is done at 1-ply). Also, it doesn't matter how the player
actually moved the checkers, this calculation is done based on its estimate
of what the best play is for all 36 rolls and the one actually rolled.

Snowie adds up all the pluses and minuses for both players and then has a
total luck factor for each game and the entire match. GnuBG goes a little
bit further and keeps it separated with all of the luck factors for my
rolls being added up and a second number for all of the opponent's rolls.

The predictive value of this number in a match is pretty interesting, as
the correlation of who had the positive luck factor and who won the match
is dramatically better than the correlation of who had less errors vs. who
won. This isn't a defect in the error calculation, more it just
demonstrates that it is far better to be lucky than good. Also, when two
players are very evenly matched their respective error rates may only
differ by 1.0 to 2.0 mppm (millipoints per move), But luck factors are
routinely in the range of 5.0 to 10.0 or more mppm in many matches. So
luck swamps skill a good deal of the time.

Also, the luck factor may not predict the result of the match as it isn't
taking the cube level into play, but looking at each game as equivalent to
a money game. So being lucky in one game with the cube on 8, but your
opponent being lucky in 4 other games that ended in a double/pass, you may
show a negative total luck factor but won the match with the 8 cube.

Gregg
--from http://www.bkgm.com/rgb/rgb.cgi?view+928

#77 · Ray Garrison

16 Feb 2014

nice to see that my idea of "ninestein" and all of it's variations is of interest to many. I think fanostein is the most interesting variant, I like the simplicity of the game and "no draws" idea. Richard M. feel free to add the variation to littlegolem....

What about a similar game, fano-streetsoccer....

#78 · FatPhil

22 Feb 2014

The FanoStein is the one with the most tension, for me, presently.
HexStein puts too much emphasis on the central hex. 4x4 would be possible, but it would be a longer game (_like that *ever* mattered in EWN!_)

Last page