Rachel Bittner
Rachel Bittner
Nothing urgent here but thinking ahead. I think we could simplify the functions in `jams_utils` a lot by creating `.to_jams` methods for the Annotation classes. This would remove the need...
Our index specifies the paths to any metadata files, but right now we write them manually when writing `Dataset._metadata` properties (e.g. in `datasets/maestro.py`). We should instead get the paths from...
Since the release of numpy 0.20, mypy is failing on our code whenever we're casting lists to numpy arrays (e.g. `cante100` line 291). I've temporarily patched this by adding `#...
Problem: When unzipping/untarring large files, mirdata is silent for a long time. Suggested Fix: Add tqdm progress bars in the `download_utils.unzip` / `untar` functions.
add extra option to validate which would allow for randomly sampling N tracks to validate (in order to support big datasets)
The `docstring_inherit` decorator isn't properly parsing multiline descriptions, and treats the second line as a new parameter. #419 removes the one multiline description from `core.Dataset` to temporarily avoid the problem,...
the current gtzan loader only loads genre and the audio, but there are other annotations available, including [tempo](http://www.marsyas.info/tempo/), [key1](https://github.com/alexanderlerch/data_set), [key2](http://visal.cs.cityu.edu.hk/downloads/#gtzankeys), [beat/downbeat](http://anasynth.ircam.fr/home/media/GTZAN-rhythm), and [metrical levels](http://www.isophonics.net/content/metrical-structure-annotations-gtzan-dataset)
After downloading files, we check that the md5 hash of the downloaded file matches what we expect before moving forward. Unsurprisingly, when the files are very big, computing the hash...
For restricted access zenodo records, we can download automatically if the user provides an access token. We can update the relevant download functions so that manual download is not necessary.