as a parenthesis, I am curious about your math typesetting tools and exeperience, here (and later on lichess blog medium). Was it tedious, and getting in the way of your sharing? or other things to share about it. I noticed upon reload some fleeting syntax behind. Was is mathML, and what kind of support for that is on here (substack, some kind of markdown+some mathML, e.g. IDK).
If you are creating new metrics, that are about another angle to view or extract some sense of existing data (through previous metrics, at another level perhaps), I am not sure there is a correct notion for that. It would come with usage, and how one can figure out some new understanding from shedding light on the evolution of a game or even about a player (or pair of players if not able to assign patterns from many games to one player given all the other pairs with that player, using that new metric).
But I think allowing for an avenue to adjust the existing metric, for perhaps various properties, is a good thing. it could even have been 1 not 2 (as long as taking absolute value).
I also now recall some conversation or email chat bot results from someone having had a long conversation (GPT3.5 I guess) that was about gradient descent methods using past values memory under some fractional feedback version, the bot was inventing something as if it would smoothen such trajectory in parameter space (say a NN weight space), and be more "aware" of the nature of the landscape surroundings (of objective function being optmized). While this was hallunications that was conflating lots of words from likely existing technical or scientific literature that might have been part of its training corpus (and I did some semantics scholar search about it, while my intuitoin, from own training, about memory in dynamical systems was not a factor of simplification of behavior), it was not real articles it was proposing as references.
But the fractional derivative and fractional "moments" if I wanted to join the statistical terminology, are part of functional (functions of functions giving numbers as output), did provide for some notions about time series "variation". This has been a while.. I could trace back (kept a category in my account about that).
I have no clue it that coefficient would provide more angles of representation into another type of ply-clocked time series (non infinite, but who cares, specially if looking at many games, this is not like about how the game classify at end position but the nature of position turn by turn dynamics).
I hope this is not a waste of reading time, and that my memory of words is not screwing the message.
I say 2 sounds more "natural" (probably looking like based on some inner product born metric somewhere) but I am not sure having a battery and exploring other values is having no discovery potential.
I also think that values greater than 1 are better as one big evaluation swing gets weighted more than several small ones that add up to the same evaluation swing. As you said, trying different values is always good, but it's difficult to asses if one value is better than another one.
makes me smile
as a parenthesis, I am curious about your math typesetting tools and exeperience, here (and later on lichess blog medium). Was it tedious, and getting in the way of your sharing? or other things to share about it. I noticed upon reload some fleeting syntax behind. Was is mathML, and what kind of support for that is on here (substack, some kind of markdown+some mathML, e.g. IDK).
Substack has Latex typesetting built in, so it's quite easy here. For Lichess I just took screenshots of the formulas and included them as pictures.
If you are creating new metrics, that are about another angle to view or extract some sense of existing data (through previous metrics, at another level perhaps), I am not sure there is a correct notion for that. It would come with usage, and how one can figure out some new understanding from shedding light on the evolution of a game or even about a player (or pair of players if not able to assign patterns from many games to one player given all the other pairs with that player, using that new metric).
But I think allowing for an avenue to adjust the existing metric, for perhaps various properties, is a good thing. it could even have been 1 not 2 (as long as taking absolute value).
I also now recall some conversation or email chat bot results from someone having had a long conversation (GPT3.5 I guess) that was about gradient descent methods using past values memory under some fractional feedback version, the bot was inventing something as if it would smoothen such trajectory in parameter space (say a NN weight space), and be more "aware" of the nature of the landscape surroundings (of objective function being optmized). While this was hallunications that was conflating lots of words from likely existing technical or scientific literature that might have been part of its training corpus (and I did some semantics scholar search about it, while my intuitoin, from own training, about memory in dynamical systems was not a factor of simplification of behavior), it was not real articles it was proposing as references.
But the fractional derivative and fractional "moments" if I wanted to join the statistical terminology, are part of functional (functions of functions giving numbers as output), did provide for some notions about time series "variation". This has been a while.. I could trace back (kept a category in my account about that).
I have no clue it that coefficient would provide more angles of representation into another type of ply-clocked time series (non infinite, but who cares, specially if looking at many games, this is not like about how the game classify at end position but the nature of position turn by turn dynamics).
I hope this is not a waste of reading time, and that my memory of words is not screwing the message.
I say 2 sounds more "natural" (probably looking like based on some inner product born metric somewhere) but I am not sure having a battery and exploring other values is having no discovery potential.
I also think that values greater than 1 are better as one big evaluation swing gets weighted more than several small ones that add up to the same evaluation swing. As you said, trying different values is always good, but it's difficult to asses if one value is better than another one.