Display help menus more promptly

[thompsonmj]

Addresses #57 (speed up help menu displays by deferring large imports until called for to execute a command).

For a brief demo of the quality of life improvement, timing runs on my laptop for displaying help menus before and after the lazy loading are shown below.

Eager loading, fresh environment:

$ time bioclip -h
usage: bioclip [-h] {predict,embed,list-models} ...

BioCLIP command line interface

options:
  -h, --help            show this help message and exit

commands:
  {predict,embed,list-models}
    predict             Use BioCLIP to generate predictions for image files.
    embed               Use BioCLIP to generate embeddings for image files.
    list-models         List available models and pretrained model checkpoints.

real    0m18.306s
user    0m0.000s
sys     0m0.062s

Eager loading, subsequently in the same environment:

$ time bioclip -h
usage: bioclip [-h] {predict,embed,list-models} ...

BioCLIP command line interface

options:
  -h, --help            show this help message and exit

commands:
  {predict,embed,list-models}
    predict             Use BioCLIP to generate predictions for image files.
    embed               Use BioCLIP to generate embeddings for image files.
    list-models         List available models and pretrained model checkpoints.

real    0m4.203s
user    0m0.000s
sys     0m0.047s

Lazy loading, fresh environment:

$ time bioclip -h
usage: bioclip [-h] {predict,embed,list-models} ...

BioCLIP command line interface

options:
  -h, --help            show this help message and exit
 
commands:
  {predict,embed,list-models}
    predict             Use BioCLIP to generate predictions for image files.
    embed               Use BioCLIP to generate embeddings for image files.
    list-models         List available models and pretrained model checkpoints.

real    0m2.612s
user    0m0.000s
sys     0m0.031s

Lazy loading, subsequently in the same environment:

$ time bioclip -h
usage: bioclip [-h] {predict,embed,list-models} ...

BioCLIP command line interface

options:
  -h, --help            show this help message and exit

commands:
  {predict,embed,list-models}
    predict             Use BioCLIP to generate predictions for image files.
    embed               Use BioCLIP to generate embeddings for image files.
    list-models         List available models and pretrained model checkpoints.

real    0m0.604s
user    0m0.000s
sys     0m0.046s

Similarly for sub-command menus.

[hlapp]

Having the help menus appear much more quickly is certainly helpful for a user. But we should still consider a cost-benefit analysis.

Obviously, there is a benefit for the user. However, the benefit is only for learning how one can control what the tool does, not for running the tool for a chosen task.

In terms of cost, it's hard to see how there could be costs for the user, but there can be and arguably are costs for developers. For example, some loss of code clarity, which is perhaps minor. If, however, this change affects (reduces) the way code completion works in widely used code editors (because they can't determine the imports anymore), then I would consider this major. And at this point in the codebase's lifecycle I would rank developer efficiency higher than a user's ability to shave off 10 seconds from the time it takes to get a help message.

So how does this affect code completion in, say, VS Code?

[thompsonmj]

The VS Code IntelliSense auto-complete recommendations are still available:

I believe this is thanks to the declaration in __init__.py of __all__ = ["TreeOfLifeClassifier", "Rank", "CustomLabelsClassifier", "CustomLabelsBinningClassifier"]

[hlapp]

I actually mostly meant code completions on the imported modules, not only the pybioclip modules.

[thompsonmj]

Ah, gotcha.

I think this demonstrates what you're looking for then:

[hlapp]

I think this demonstrates what you're looking for then:

Not quite. But I notice that it's only the open_clip module that would now be hidden away (and that presumably is causing the delay otherwise, due to its dependency on Torch etc?). Will it still autocomplete on open_clip?

[thompsonmj]

Here's open_clip autocomplete working:

And yes, open_clip as well as the torch imports in predict.py were biggest contributors to the delay.

[hlapp]

But your screenshot is with a direct import. Of course that works, that wasn't my concern.

[thompsonmj]

How about this?

Where list_pretrained_tags_by_model is an oc method.