What does it mean if a function perfectly describes a data set? One version or view is that that a raster-like bounded data image, perfectly stated as a function (there’s that function I saw on youtube that draws any raster image) could have
predictive irrelevance, unless there was a “position” or “environment” operator that showed which way the data would drift if some equation were applied to it, as an instant function, or time applied tothe drift, or a new time-styleform applied
to the fully-predictive equation (like the raster one).

So just thinking about it, “What does it mean if a function perfectly describes a data set?”, if, big if, the data set is based on observed things like science, It could mean that the function has some non-zero probability of having “leading
information” as to what actual thing the dataset is!

You might be able to have probability above chance ((at a finite, or with sufficient math, perhaps nonfinite) at a list of possible things something could be) like something is more likely to be physically sourced in time series data if it had a poisson
distribution, or it might be able to speculate that something with a normal distribution was made up of many samples of something that could be multiply recombined.
So, a function that defines a dataset in its entireity, at least with numbers derived from measurement, might contain clues as to what the dataset is. Now I perceive there are an arbitrary nonfinite number of equations that will describe any finite
dataset, so the idea of winnowing out “what it might actually be” from chance seems peculiar, but if you make a big list of every equation that has previously been used to predict something at the scientific literature, and then you use that short
list of 1 million (or more) equations to see if any of them are partially, or quantifiably partiually expressed, like among the million stats models to screen the data-containing function for, a poisson distribution might be found ob being “leading
information” of the function that precisely produces all the actual data.

Then you could do an experiment, getting more data (than that described at the all-data-descriptive function), to verify or refute that the poisson distribution was actually a valid statistical shape, or possible methodology, at the one-function data.
Poisson treats things that change over chronointerval. So you could do something non-time dependent, that does not change chronointerval, and still see if it changed the data in a way that reinforced or supported, or even refuted the previous “fit”
of thinking the data was a poisson distribution.

So this idea that you might be able to find, or dredge up, something (like a numeric relation, possibly a statistics model), that at a previous million scientific publications, had predictive value somewhere sometime, and systematically find it in an
equation or function that precisely, even like raster-equation, produces all the fixed and actual (measured) data could be possible.

This compare it to the big set of all previous valid models, ever, thing: Brings up the idea of something new to me, which is that there are big human knowledge sets you can compare everything to, then do instantaneous math to solve. The possible
efficiency or new math of value here is that when you solve for two equations (Function one: function might raster-scan fully describing all data. Function two: a big pile of math instantaneously represents all previously useful statistical math, like T
value equation, Z value equation, poisson distribution; a million others)
This overlaying (solving) of the two functions can be instantaneous, where dolving the two instantaneous equations is morer rapid than something that iterates; that makes it more rapidly solvable with computers. Also, the generation of fresh equations
comparing the stats/numeric methods “periodic table like thing”could have optimizable completely new equations that describe things, or computer algorithms if the mathematician or programmer felt like doing the new knowledge generation; (those new
equations improves velocity of computation and likely extends capability as well)

The benefit here is that a person has some data, then on screening the 1 million stats/numeric methods “periodic table” this program tells them what its about: like, “periodically repeated poisson and log distribution found” often describes
repeated cycles of (perhaps seasonal or daily growth). Perhaps an improved piece of software suggests that the data, on periodic-table-style screening could suggest to the human that the driver and source of the data might be a time of day at a sytem
with moving components.

Unbeknownst to the software or the human, the software might have been looking at trolley ridership. The program found drivers and numeric methods sources without referring to the actual matter or energy composition of the thing being analyzed.

If any of the actual existing matter-based structure is available to the human, then of course the human can look at that, devise an experiment, and indeed change the trolley fare on time of day, to see if it affects ridership. The thing is that using a
million previously predictive equation set (periodic table function comparison system), these kinds of things like suggesting trolley passes, or rather adjustments to the equation with a predicted mechanism that is equivalent to introducing a trolley
pass, can be done and tested without a human-generated hypothesis.

The software could find and describe areas of the overlapping (function 1 function 2) equations, where slight changes are most likely to have large measurable effects. The software could suggest: “find an item which can be rescheduled, if it changes
the poisson-distribution-like-character of the fully-describing raster function then the poisson description is more likely to be an actual driver.”

Really impressive software could come up with winnowing the ease of experiment, that is suggesting the easist possible change that validates or refutes a periodic-table-of-one-million-numeric-methods numerically autogenerated or the fresh human (or
various AI) generated hypothesis.

Better software could winnow tests of these hypothesis with suggesting experiments that are the equivalent to giving out complimentary bus passes and finding out how the new data changes, as compared with the more effortful rescheduling of trolleys, or
the effortful moving of entire trolley stations; any of these experiments would modify what appeared to the software to be a poisson distribution, so with that modification the idea that a poisson distribution is a descriptor/driver of the phenomenon
issupported.

The software has suggested things used to verify the poisson (chronological interval sequenced) distribution being an authentic attribute of the system. So the result of the winnowing and experiment would be a new system predictive equation, that has
generated its own deductive “looks like an actual thing” component, by finding a plausible and successfully tested deductive-able from the periodic-table of numeric methods list that got simultaneous-function solved. Note humans and AI can of course
think of complely new things outside and/or possibly addable to the numeric methods periodic table. It is just that this is software that finds both standout bases for deduction as well as generates more accurately predictive equations.

(more notes that came before some of the previous writing)

(about the library of a million or more previously published numeric relationships) It is primitive thinking on my part, but it reminds me of the periodic table. You can scan any physical object against it and get a match, and that match (elemental
composition) then predicts what the thing might do, and what it can be turned into. So layering an equation, possibly a function, that happens to produce a poisson distribution onto a periodic table of 1 million statistics models) makes it say, since it
is an observation of an actual thing it might be: “poisson distribution”. Then like an element on the periodic table, “poisson distribution” means, that, besides testing new hypothesis and/or validating a model’s equations:

You can do things to it to get predictable, possibly technologizable effects or experimentally predictable effects just like you could with an element: like halve sampling rate yet still have numeric adequacy to make p<.01 predictions; or, on finding
that a normal distribution was present, that you could double sampling rate to measure a new value farther out along a standard deviation value, like a really high number or a really high spectral emissions band event. So a variation on the million
numeric method thing that finds sources and drivers of actual data, including new synthetic deduced thingables, could be used to suggest different measurements that that improve phenomena investigation, basically kind of like this suggests better optics:
see rarer events, see bigger or smaller events, possibly even trace event contingencies with an experiment/new measurement on the system. That has technological applications as well, It is kind of obvious to a person, but if you say, “measure twice
as many leaves, you are likely to find a new biggest one” and you want to find a plant with big leaves so you can breed it to be a better agricultural crop, then the software suggesting doubled frequency of observation has a cognitively comprehensible
technology benefit.

Also the math of a static function descriptor of the data (there are much better ones than the raster equation as a function that can describe an entire data set, but it was mentioned); when overlain, or instantly-function computed aka equation solved at
a million element instantly solveable, definitional eqaution, (is it possible to make a really big equation that simultaneously contains many numeric method equations in it?) is possibly much faster than some other methods.

That high instant solvability version of the equations, among them the one that generates all the actual data, could make it so some questions about the data, and/or the source/structure/drivers of the data are computable as an effect of the overlaying
or simultaneous solution of two equations; that is it theoretically instant, as compared with the duration of iteration at a computer-software derived solution.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)