BUG: predict empty file handling

[nhoglen]

Describe the bug
predict breaks when converting segments to crowsetta seq if no segments were found in the file.

To Reproduce

vak prep check_this_recording.toml runs fine.

vak predict check_this_recording.toml runs until vak.core.predict - INFO - converting predictions to annotations then throws this error:

Traceback (most recent call last):
  File "C:\Users\Nerissa\anaconda3\envs\vak-env4\Scripts\vak-script.py", line 9, in <module>
    sys.exit(main())
  File "C:\Users\Nerissa\anaconda3\envs\vak-env4\lib\site-packages\vak\__main__.py", line 48, in main
    cli.cli(command=args.command, config_file=args.configfile)
  File "C:\Users\Nerissa\anaconda3\envs\vak-env4\lib\site-packages\vak\cli\cli.py", line 49, in cli
    COMMAND_FUNCTION_MAP[command](toml_path=config_file)
  File "C:\Users\Nerissa\anaconda3\envs\vak-env4\lib\site-packages\vak\cli\cli.py", line 18, in predict
  File "C:\Users\Nerissa\anaconda3\envs\vak-env4\lib\site-packages\vak\cli\predict.py", line 50, in predict
    core.predict(
  File "C:\Users\Nerissa\anaconda3\envs\vak-env4\lib\site-packages\vak\core\predict.py", line 244, in predict
    seq = crowsetta.Sequence.from_keyword(
  File "C:\Users\Nerissa\anaconda3\envs\vak-env4\lib\site-packages\crowsetta\sequence.py", line 382, in from_keyword
    labels) = cls._validate_onsets_offsets_labels(onsets_s,
  File "C:\Users\Nerissa\anaconda3\envs\vak-env4\lib\site-packages\crowsetta\sequence.py", line 224, in _validate_onsets_offsets_labels
    raise ValueError('must provide either onset_inds and offset_inds, or '
ValueError: must provide either onset_inds and offset_inds, or onsets_s and offsets_s

I did a little bit of digging and determined that audio segments without any calls are tripping up the crowsetta conversion process. Specifically, at line 239 of the vak predict.py file, there is a check for empty segments that is going awry because labels is turning up an empty list rather than a None object.

if labels is None and onsets_s is None and offsets_s is None:
                # handle the case when all time bins are predicted to be unlabeled
                # see https://github.com/NickleDave/vak/issues/383
                continue
seq = crowsetta.Sequence.from_keyword(
                labels=labels, onsets_s=onsets_s, offsets_s=offsets_s
            )

Inspecting the variables at this step shows that onset_s and offset_s are None as expected, but not labels.

(Pdb) print(len(labels))
0
(Pdb) print(len(onsets_s))
*** TypeError: object of type 'NoneType' has no len()
(Pdb) print(len(offsets_s))
*** TypeError: object of type 'NoneType' has no len()

Expected behavior
In my previous vak install (version 0.4.0, I think), this ran to completion and generated a full set of predictions.

Desktop (please complete the following information):

• Operating System: Windows 10
• Versions: vak 0.8.0, TweetyNet 0.9.0, crowsetta 3.4.1

Additional context
I will email the files to reproduce; the zip is too big to upload here.

[NickleDave]

Sorry you had this issue @nhoglen and thank you for this very thorough bug report.

I'll know for sure when I test with the data but my hunch is you have correctly identified the issue; we do not correctly handle the case where there are no predicted segments because we don't look for the empty list

This was supposed to be fixed by #394 but clearly it wasn't!

Could you please share the data some other way? I wasn't able to open the sharepoint link. Will ask by email too but wanted to reply here (and thank you profusely for making this easy for me to fix 😼 😁 )

edit: Box link you sent worked, nvm!

second edit: just to cross-link, this was originally posted on the forum: https://forum.vocalpy.org/t/error-in-last-step-of-generating-predictions/54

[NickleDave]

🤔 I'm not able to replicate this bug -- vak predict runs fine for me on Linux using the files you provided.

Could you please reply with the following files attached?

• the .toml configuration file you're using
• the output of running the following command in your conda environment: conda env export > environment.yml
• as well as conda list --explicit > spec-file.txt (also in the conda environment you are using when you get this bug)

I'm attaching the .toml file I used as well as the csv of annotations that was produced, in a .tar

predict-results.tar.gz

fwiw I have the same versions of vak, tweetynet, and crowsetta installed. I can try making an env on Windows with conda files just in case for some reason the platform is the source of the issue.

Thank you! 🙏

[nhoglen]

Oops, sorry I missed this -- I'm not used to looking out for things on GitHub.

I have found that getting the environment set up right on Windows takes some finagling, so I would not be surprised if there are weird version issues baked in there.

The attached zip has the listed files. environment-etc.zip

I'm happy to keep poking at this and will keep a closer eye on GitHub/figure out how to get it to send more notifications.

[NickleDave]

Thank you @nhoglen! That helped.
It was an option in the TOML config file that I was missing, the min_segment_dur.
Once I added that to the config I had, I was able to replicate your error.

I now also get the same exception that you first reported on the forum:

Traceback (most recent call last):
  File "/home/pimienta/miniconda3/envs/vak8-env/bin/vak", line 10, in <module>
    sys.exit(main())
  File "/home/pimienta/miniconda3/envs/vak8-env/lib/python3.8/site-packages/vak/__main__.py", line 48, in main
    cli.cli(command=args.command, config_file=args.configfile)
  File "/home/pimienta/miniconda3/envs/vak8-env/lib/python3.8/site-packages/vak/cli/cli.py", line 49, in cli
    COMMAND_FUNCTION_MAP[command](toml_path=config_file)
  File "/home/pimienta/miniconda3/envs/vak8-env/lib/python3.8/site-packages/vak/cli/cli.py", line 18, in predict
    predict(toml_path=toml_path)
  File "/home/pimienta/miniconda3/envs/vak8-env/lib/python3.8/site-packages/vak/cli/predict.py", line 50, in predict
    core.predict(
  File "/home/pimienta/miniconda3/envs/vak8-env/lib/python3.8/site-packages/vak/core/predict.py", line 243, in predict
    seq = crowsetta.Sequence.from_keyword(
  File "/home/pimienta/miniconda3/envs/vak8-env/lib/python3.8/site-packages/crowsetta/sequence.py", line 382, in from_keyword
    labels) = cls._validate_onsets_offsets_labels(onsets_s,
  File "/home/pimienta/miniconda3/envs/vak8-env/lib/python3.8/site-packages/crowsetta/sequence.py", line 224, in _validate_onsets_offsets_labels
    raise ValueError('must provide either onset_inds and offset_inds, or '
ValueError: must provide either onset_inds and offset_inds, or onsets_s and offsets_s
(vak8-env)  ✘ pimienta@pop-os  ~/Documents/data/vocal/bug-report

It does look like this is in fact a bug I introduced when I added a new feature here:
8c6aee1
where for some reason that's not clear to me now I changed this function to return an empty string when there were no predicted segments after post-processing (e.g. with a min. segment dur):

vak/src/vak/transforms/labeled_timebins/functional.py

Line 191 in 8c6aee1

return '', None, None

whereas the same condition would have caused the previous version of this function would have returned all Nones:

vak/src/vak/labeled_timebins.py

Line 441 in 25a0c8a

return None, None, None

Not sure if changing it back will break any other changes I made--will find out when I test--but my plan is to change it back for now and release a bugfix version.