add cli command that generates an empty config file

[Luke-Poeppel]

The Auto-Annotate tutorial suggests downloading a given config.toml file. But is there currently a tool for creating a template config file, e.g. something like vak make /test_a/config.toml --train_mode True or vak make /test_a/config.toml --predict_mode True with empty values? Maybe this exists and I missed it in the source –– as such, I'm not sure if this is a docs or feature request 😄

[NickleDave]

Hi @Luke-Poeppel
No there is not currently a way to generate a config file through the cli, but yes I have thought about it and agree it would be nice to have.

I haven't raised an issue about it though, so now this is that issue, thank you

I guess I would favor the command vak init with flags '--prep', '--train', '--learncurve', etc

Here's how poetry does it (for future reference):
https://github.com/python-poetry/poetry/blob/b753aaf4c3c08ef0e54941a6616fe318fdf4f6e4/poetry/console/commands/init.py

[NickleDave]

a couple further thoughts on this:

• I think I would rather have a command that simply generates an "empty" config file
  • instead of having a series of interactive questions.
  • In my experience it's easier to just open up the file in a text editor and cut and paste things by hand.
  • Some people who live in the terminal might use an interactive Q&A at the prompt, but most people wouldn't (I think)
• would prefer an explicit name for the command
  • init was a bad idea (mine) because we're not initializing anything
  • don't want to clash with config because that would be natural for configuring the behavior of vak itself
  • so I guess a command name should be something like makeconfig or maketoml, prefer the former because not everyone knows what a "toml" is.
  • Or, more verbose: $vak configfile make train
• a command like vak configfile make train implies we now have multiple sub-parsers though
  • an additional point to consider is that in my mind the cli should stay as simple as possible;
  • the goal is for anyone who needs to do anything more complex to be able to do so with the actual API of the library, not to have a super-fancy cli
  • but still worth the extra cli functionality here, because it is annoying to not just have config files at hand

[NickleDave]

Coming back to this.

Noticing during tutorials that there's a lot of cutting-and-pasting of links.
E.g. a lot of the options in the eval and predict section are basically fixed links that come from a results_dir.

Would be nice to be able to say

vak config predict --results_dir ./path/to/train_results/results_20221312_100439

and get a config with all those options already filled in

[NickleDave]

• do this after adding dataclasses in ENH: add dataclasses to represents results directories, e.g. TrainResults and LearncurveResults #449 since it would be easy using those dataclasses
• If we end up doing this, add it to the auto-annotate tutorial!