vak
vak copied to clipboard
vocalpy
A neural network framework for researchers studying acoustic communication
As discussed in #667 the current dataset classes conflate the underlying source data (e.g. audio files, array files with spectrograms, annotation files). This also results in hard-to-reason-about logic inside the...
- [ ] e.g. to use 'model' option not 'models', see: https://github.com/yardencsGitHub/tweetynet/issues/220
An API reference was added in #644 to fix #441 but there is definitely a lot of cleaning that can be done.
I didn't add a test in #636 because I wanted to get the fix out. We should - [ ] add a unit test that asserts we get expected output...
- [ ] make an initial version of `wences` that just combines clips using e.g. a first-order Markov model - [ ] use this to generate fake spectrograms e.g. with...
those added in #621
as discussed in #472 > Change what we return so that we have access to the time bins vector t in a batch when we call eval - [ ]...
In #611 we added the ability to control how `audio.to_spect` parallelizes creation of spectrogram files by calling `dask.bag`. A user can do this with an option in the [PREP] section...
to replace `vak.models.models.from_config` etc. should look something like this ```python # vak.models.get from .base import Model def get(name: str, config: ) -> Model: # actually an instance of some model...
as in #605 - [ ] vignette: adding a WindowedFrameClassificationModel - [ ] reference: abstractions for models, base class + decorator; goals and design decisions Should adopt language + snippets...