David Nicholson

- [ ] use `vocalpy` classes to convert audio -> spectrograms if specified - [ ] make a `vocles` dataset and save that by calling `to_csv` if possible prefer clearly...

discuss: how to fix size of dataset without WindowDataset abstraction?

1

#555 proposes to remove `WindowDataset` and instead have the ability to randomly crop each vocalization. One reason to have the `WindowDataset` abstraction is it lets us precisely fix the size...

ENH: Deprecate WindowDataset

1

Currently training is tightly coupled to the `WindowDataset` abstraction. A clear drawback of this is that it prevents us from training in any other way, e.g. feeding in an entire...

#555 proposes to deprecate `WindowDataset`. One good thing about the `WindowDataset` abstraction is that kind of gives us data augmentation and thus translational invariance "for free", because a network can...

ENH: `eval` should have option compute metrics with and without clean-ups

1

as in paper. Should add same options as in predict config and then adapt from this script: https://github.com/yardencsGitHub/tweetynet/blob/master/article/src/scripts/run_eval_with_and_without_output_transforms.py

Improve train/test(/val) split functions?

2

Making this into a general issue / discussion of how to improve these functions. Will collect thoughts + ideas from other issues here (to close those and consolidate).

something like sklearn https://scikit-learn.org/stable/faq.html#new-algorithms-inclusion-criteria to avoid a Pelican-like situation

To simultaneously be more specific (because there are other .csv paths) and be more general (because a dataset may not always be a .csv) Probably not a good idea to...

related to #446 we do already have a base VocalDataset but it's basically just used for prediction there should be something like a base `Dataset` class similar to the hierarchy...