I guess I'm just an outliar, only 1800 FIDE.
@grandstudies said in #31:
> I guess I'm just an outliar, only 1800 FIDE.
Are you regularly playing OTB?
What comes next ... stretch the dots out until it makes nearly a straight line. Stop the rating at 6,150.
If we had a rating limit of 6150, it would help every player show how much they’ve improved over time and make it easier for everyone to find their perfect match on the chessboard!
@RookyBeach said in #29:
Yep exactly on same page and it means we need to see the full model
I do appreciate the extra explanations.
Even the nonlinear family approximation was pleasant to hear from here
Still it is not because it is piece wise a given simple function family that it is inherently more flexible or expressive than a global closed formula or smooth family of function
Think polynomials
In the end though you are right to raise that point it means we should check. But that would be the same problem with a one piece formula with same number of parameters as the number of pieces boundaries
I think this calls for seeing the complete model formulation with the pieces parametrized
The figures can only convey pairs of dimensions in this many pools intersections and missing data analysis experiment
Well that would be interdesting
Not sure it would be needed for personal approximation of own rating
Depends on own precision or confidence needs
Just saying It might be further research
And yes number of parameters is not really a measure of complexity in the expressiveness sense
But that is true piecewise or not
Need to see full function space been optimized
Expressiveness is about the ambient function space « size » or shape why I find the natural word flexible more evocative
But in ML I think the word complexity is already taken for the combinatorial algorithm stepping complexity
Expressivité is pointing more to the topological (open set basis not graph topology only) complexity
I come from modeling using dynamical system language and there complexity was about the geometric one how many bumps to describe a function
Smooth or constant function being less complex than a fractal function
Cantor stairs. Bifurcation cascades etc
Sure if one had to define finite problems that might be solved by algorithm schemes and finite size parametrization
Then I would think both notion of complexity would be in some monotonic increase increasing relation not identity and not likely linear or smooth or closed form computable
Expressive enough to fit as close as possible any function that the law of probability of the phenomenon being sampled representativeky by the data, any complexity of that function as there can be in some convened big enough function space
I might delete this when I get a real conputre
I can’t maneuver on this cell this time was off topic reminiscing on past things that having some one who had some math language overlap here
How about blitz ratings? Do you have a separate analysis about those? Or are you aware of any?
Basically I am in the 2500-2550 range in blitz here and curious what FIDE rating it would translate to. I have a FIDE rating, but play very little, so it is probably not reflecting my strength these days.
Factor in a time differential to convert a rating. That way with less time our rating should drop and with zero time the rating should be zero.
I think one trouble with Fide ratings compared to national ratings is troublesome as the fide ratings are often lower in absolute numbers than the national rating mainly because of a lack of played fide rated games in the past. The fide ratings are often underrating players who very rarely have played fide rated tournaments in the past, and improved later in their national rated tournaments.
I am pretty sure that a 2100 ecf player is stronger than a 2100 fide rated player, but your table suggest the opposite.
If a 2100 ecf player has 2100 fide rating, it often is because he played a fide rated tournament when he was weaker, and then rarely played any fide tournments to improve his fide rating (Old rating + a few new performance ratings lead to the ratings not appreciating fast enough).
It is really difficult to improve the fide rating of lets say 1800 to 2000 if you only play 1 fide rated tournament of 7-9 rounds every 2-3 years, while your established ecf might already be 2000, where you have played 12-15 ecf games each year.
Thus, ye, the number in the database suggest fide is stronger than ecf, but you can already see that indian fide rated players (who hardly have played fide rated tournaments) are severely underrated when playing in international tournaments and get matched against european fide rated masters. you often can say a indian fide rated non-titled player is 100-200 fide elo stronger than a regularly european fide rated GM.
you can regularly see european/american GMs lose fide rating points when they go and play in Dubai for instance, often against indian players with underrated fide ratings due to not having played many fide rated tournaments.
and for ratings at lichess: I am a ecf 2100+ rated player, and struggle in rapid against 1800 ecf/fide/dwz players in lichess in rapid or classic, while I do not OTB.
ok, rapid isnt classic, but it should not be that off the mark.
The explanation why I struggle in lichess and lag like 300 rating difference might be:
a) cheaters
b) I do not take the games seriously, often switching to youtube while my opponent think
c) difference in arena tournament ratings and open challenge ratings. Really, the lichess tournament player at lichess are weaker. Maybe berserker distorts ratings, and also - if you do not play the first move, you lose rating points in a lichess tournament, while in open seek challenge the game just gets aborted.
d) time controls: there are 120 minute games as classic and 25 minute games, all having the same classical ratings (and with rapid similar: the time control between 8 mins and 25 mins are all "rapids", and if you play 1.5 min + 35 sec, you are still "rapid")
with point b) I am not sure, as that should be with my online opponents on average the same for them, too, they might not take it as seriously as OTB, too.
@F-35_Raptor said in #15:
> ok, but thats still innacurate, I could easily reach 2400 classical if I wanted to but lets leave that
They say in the last part of the blog that past 2000 FIDE there is no accurate way to estimate online classical to FIDE, because there aren't enough high rated online classical players. 2400 classical on lichess is only 20 players with low deviation. 2300 classical is only 70 people.
That's why I don't really like this estimation system. There aren't enough sample points.