Various Tool(s) to "Create" configuration Files

[FencerDave]

For different levels of user, we should have a variety of options for the configuration files...

• For advanced users, we already have comments/instructions in the Config files themselves with instructions on how to write them.
  • These can always be expanded.
  • We discussed having explanation ReadMe.md files with more detail to guide Config writing.

For others, we have a lot of options (requiring various amounts of work) 1) "Kitchen Sink" option based on Data (for Tform Config):

• Load up the user's data, inspect the columns, and generate ALL transformation commands that would work with their data set. (For instance, cannot Log() if has negative data...)

• Then the user can manually delete any that they don't want to try.

  2) "Runtime Limited" version of Kitchen sink could be implemented, where the user specifies a run time (4 hours, for example), and the loop will continue until that time has elapsed, with however many transformations are done in that time.

  2b) Similar Manual-ish version could make a kitchen-sink-like estimate but simply cut off the list after a certain number of iterations.

  3) "Priority by Inspection" : We could use a simple (existing) linear-correlation algorythm (Numpy has some right?) to do an early correlation between each column and Linear, Log, Exp, Power-series, etc. data, and then put THOSE transformations as the first to try. This is pretty "A-Priori" so maybe unfair, but it might be the quickest path to good results!

4) "Priority by Meta-Learning" : This might be the best End-Goal... Start out with random transformations, then iterate the loop until the program "LEARNS" which transformation/column pairs are associated with the best models. Eventually, it will find the best ones to use on its own!?!?!

[FencerDave]

(Again, please feel free to suggest additional ideas, or comment on the value/priority of the above!)