NickleDave
commented
3 months ago Thinking out loud: I guess most tools are pretty consistent about how they save so it would have to be someone working with the exported annotations that corrupts the files.
It would be good to know if anyone else had cases where pandera caught some problem with annotation files, and that was helpful. I think most of the validation errors I've gotten have been because of a mistake I made (e.g. building a simple-seq annotation file "manually" with pandas from some one-off annotation format, then trying to load it with crowsetta)
Maybe we just need clearer error messages instead of strict validation
Our using pandera to validata dataframes really adds to the number of things that get installed when you install crowsetta, largely because pandera depends on pydantic
This makes it it more likely that some change upstream will impact people that just want to use crowsetta so their own library can parse annotations, see for example https://github.com/kitzeslab/opensoundscape/issues/1017 and https://github.com/vocalpy/vocalpy/issues/173
I recall looking at "pure Python" libraries for validating dataframes before, I wonder if there's one we could vendor to avoid. Like typedframe maybe