Are variants "real" chess?

Oh, Thanks, Dude.

@polylogarithmique said in #46:

@piazzai for Glicko I don't know. For Elo, the details can be found there en.m.wikipedia.org/wiki/Elo_rating_system
under section "theory".
@EmaciatedSpaniard said in #47:
There is such a formula in this document on glicko ratings. See page 10 and look for the Standard winning expectancy:

http://www.glicko.net/ratings/rating.system.pdf

Thank you very much for this. I didn't know about the formula. The hurdle I see to relating this to my model results is that the formula quantifies the expected score, which is the probability of winning plus half the probability of drawing. I can only speak to the odds (and, with some assumptions, of the probability) of winning. But many different probabilities of winning map to the same expected score.

@JKS369 said in #1:

While I like variants (atomic and racing Kings) I have gotten some criticism about them not being "real" chess. Are variants "real" and do they help improve our standard chess skills?

Variants are not real chess (except Chess960). They don't help you to improve chess. Just you can play variants if are interesting for you. Example I love variants and I'm playing) Good luck!

@piazzai I am not sure I understand your hurdle. Can't you import all the games where player A's rating is player's B rating plus 100 (say), add one for every game A won, add 1/2 for every draw game and divide by the total number of games? This would give you the empirical average, which should in principle match the expectation given by the formula?

I see the problem now, you put draws as an undesirable outcome.
Draws must be excluded if you ever wanna translate your numbers back to something meaningful.
Can’t make assumptions correlated to rating if you dilute the numbers affecting said rating. I thought that’s the point of choosing similar ratings, so you can discard draws.
@piazzai

@polylogarithmique said in #54:

@piazzai I am not sure I understand your hurdle. Can't you import all the games where player A's rating is player's B rating plus 100 (say), add one for every game A won, add 1/2 for every draw game and divide by the total number of games? This would give you the empirical average, which should in principle match the expectation given by the formula?

Of course, that is just a few lines of code. I was stuck thinking about how the formula relates to model predictions but you are not asking about that. You are asking about how it holds up with empirical scores.

Here is the comparison: https://raw.githubusercontent.com/piazzai/chessvariants/master/images/expectedscore.jpg
The empirical score becomes a bit less reliable toward the left and right extremes because there are fewer games with such a large difference between the players. The computations are based on any game, variant or standard, at any time control.

So we see that the formula tends to over-estimate slightly the chances of the higher rated player to win :)

Depends on what you call "real" chess, most variants require planning just like "real" chess or have the same rules as chess for example Chess960.