David Nicholson

@kalleknast just icymi I did release a version 0.8.0 today that includes this https://github.com/vocalpy/vak/releases/tag/0.8.0 You should be able to `pip install` now and install off `conda-forge` sometime tomorrow. Thanks again...

Awesome, glad to hear it

Picking this up again. The order of operations needs to be: - first, implement new Model class/decorator as in #536 to fix #362, because the new engine will expect a...

Closed by #605

Similar to this `CharMap` class: https://github.com/jeremyfix/deeplearning-lectures/blob/b3862d6dd1af45bea1a99f9b26a0c8baa1520422/LabsSolutions/02-pytorch-asr/data.py#L115 We don't need all these methods but seeing someone else has had the same idea makes me think it would in fact be good...

Not going to do this for now, as discussed in #651 and issues linked there. Closing

> A clear drawback of this is that it prevents us from training in any other way, e.g. feeding in an entire vocalization (e.g. a birdsong bout) as one sample...

One issue with implementing this is that, to convert network outputs to a string of labels in the `Model._eval` method, we actually use a different function from what we use...

I got as far as converting `labeled_timebins` to transforms and started to add `post_tfm` to `Model` but now I see an issue that I missed. To be able to just...

Update on this one: Question for version 0.x is how to add this. Doesn't matter too much since there's really only one model but version 1.x will need to be...