James Owers

icon company University of Edinburgh @cdt-data-science & @BayesWatch icon location Edinburgh icon location At the Centre for Doctoral Training in Data Science (@cdt-data-science) at the University of Edinburgh. July 2016 - Working on Music and Neural Networks

Results 32 comments of James Owers

Automate documentation creation

Actually- look into NBDev in place of binder, just quickly! https://github.com/fastai/nbdev https://www.youtube.com/watch?v=9Q6sLbz37gk&ab_channel=JeremyHoward

Automate documentation creation

This also looks awesome! https://github.com/mwouts/jupytext

Before release: DQA of the released data!

I think it will be best to add this as a notebook to the ACME repo too. Keeping issue here. Do this in conjunction with #129

Before release: DQA of the released data!

Actually, I'd like to keep this. I've done some basic looks at the data, but haven't addressed specific things in the description.

Sort out the pip install and release on PyPI

Started the process with #142

Trainers and formatters

The thing to consider first is whether to use pytorch lightning

Handle case of zero length notes created in formatters

We decided to force all notes to be at minimum length 1. This is because the command format is complicated by length zero notes i.e. note off followed by note...

Implement --baseline in train and eval scripts

This is nice, i think a priority for camera ready, but not ACME v1.0 release.

Implement --baseline in train and eval scripts

Change of plan, we should remove this flag in favour of just documenting how to reproduce results.

Make sure we attribute and credit correctly

I'll make sure these are all added to the relevant README.md