Output Structure

[pearsonca]

I'm recording my notes here on output file structure as I pick through the homework.

I definitely agree with @TimothyWillard's point in #257 and I intend these notes to be useful for making an implementation that makes the objective in that issue easy to accomplish (and the equivalent in other languages, since we should be using different tools for different tasks).

• I agree with having a default run_id, but I think the default should work in a different way: it should always be "the same", instead of defaulting to time. As in, if I run a config with no changes a little later, I should get the same result (even if the simulation is stochastic). If I change the config, I should get a different result with the default run_id assumptions.
• the config input file should be snapshotted into the output directory somehow (separate concern: probably along with something like the relevant versioning info for the repository, flepimop, maybe hash value of input data, etc - not sure how we currently manage that kind of manifest information, but where we're using version control, we ought to just rely on that rather than snapshotting ourselves to death)
• the directory names are bit mismatched from meaning now, and a bit more terse than necessary (ideally users won't engage with them directly, but some inevitably will and of course developers will)
• on the flipside, the names seem more verbose than necessary. e.g., why repeat seir in the file name? the path includes seir - use that as information. including extraneous items makes it more difficult to regex out other meta data (i.e. the slot.block.iteration bits). Apparently: this verbosity is associated with prior version of arrow reading information from file name, but now that's deprecated in favor of reading those features from the path. We should update to that version (and then my complaint about the verbosity would also be solved)