Update output command-line description

[bittremieux]

Fixes #241.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.54%. Comparing base (3c2d3f5) to head (ac60313).
Report is 109 commits behind head on dev.

Additional details and impacted files

@@           Coverage Diff           @@
##              dev     #276   +/-   ##
=======================================
  Coverage   89.54%   89.54%           
=======================================
  Files          12       12           
  Lines         918      918           
=======================================
  Hits          822      822           
  Misses         96       96           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[wsnoble]

How does Casanovo handle model file naming when fine-tuning a model? In this case, it seems like --model will be used to specify the name of the input file AND the output file. Problem! Do we need separate --model-in and --model-out options? Alternatively, is there some scheme whereby we name the trained model with the epoch number in the filename? If so, this should be documented here.

[melihyilmaz]

How does Casanovo handle model file naming when fine-tuning a model?

By default, checkpoint files are named in the epoch=40-step=2550000.ckpt format regardless of the input ckpt filename. This is the case for both fine-tuning and training from scratch.

[wsnoble]

OK, so this is obviously bad, because if I train two models from the same working directory, the files will overwrite one another. We should change this to be <model-name>.epoch=40.step=2500000.ckpt, where comes from the --model parameter. This behavior should also be documented.

[wsnoble]

I just ran into another problem which needs to be fixed, probably on this PR. It seems like Casanovo is trying to do something smart if I specify an --output option with a period in it. Anyway, the result is that if I do something like --output foo.bar, then the output is foo.log. Worse still, it seems like we have some logic in there that prevents it from overwriting existing files, but it does this silently. So I ran a big sequencing run using a name like "foo.filtered" but because "foo.mztab" and "foo.log" already existed, Casanovo simply ran to completion but did not produce any output files.

As we discussed, we should add an --overwrite flag that allows users to ovewrite existing files. Otherwise, Casanovo should halt with an error if it tries to overwrite an existing file. Ideally, when sequencing it should check for file existence before starting the run, rather than at the end.

The argument to --output should be used as the prefix for all output files, with no attempt to parse it. The argument may contain directory names, but we will make no attempt to verify that they exist.

[bittremieux]

Superseded by #372.