0.3.1

New in 0.3.1:

• Allows passing file-like objects in addition to strings in most loaders.
• Introduces @core.coerce_to_string_io and @core.coerce_to_bytes_io decorators
• Moves boilerplate code in inherited Track classes to core.Track.__init__
• Moves dataset metadata from a module-level LargeData objects to Dataset._metadata
• Reintroduces strict black code formatting in pull requests
• Runs mypy as an additional test in pull requests
• Fixes typos and formatting issues in docs

0.3.0

New Loaders

• AcousticBrainz Genre
• cante100
• IRMAS
• Saraga Carnatic
• Saraga Hindustani
• Tonality classicalDB

Features

• Changes top level API - datasets are now loaded with mirdata.initialize('orchset')
• Adds a Dataset object
• Renames load --> load_tracks
• Adds dataset.choice_track()
• Moves Track and MultiTrack base classes from mirdata.track to mirdata.core
• Extends index structure beyond tracks and adds version as mandatory top-level key
• Introduces of datasets submodule
• Introduces of Annotation classes for different time-series data types
• Improve documentation
• Introduces support for remote indexes for big datasets
• Specify license in each dataset
• Config flags for automatic full-dataset testing each week (#337)

Bug fixes

• #79 RWC beat parsing
• #311 Beatport Key fix index
• #326 Maestro downloading twice breaks
• #328 RWC metadata not parsing
• #329 Download message was not showing path
• #341 Support datasets with no metadata
• #346 Mridangam fix
• #347 Orchset unzip error
• #348 and #306 file download encoding utf8
• #351 Tinysol test_full_dataset fix
• #357 Cante100 partial download fix
• #374 adding bibtex to AcousticBrainz

0.3.0b1

New Loaders

• Saraga (no multitrack)
• IRMAS
• cante100
• Saraga Carnatic
• Saraga Hindustani
• AcousticBrainz Genre
• Tonality classicalDB

Features

• Updated Dataset object to be extended by every loader
• Switched annotation datatypes to dedicated classes from NamedTuples
• updated index structure to support more than tracks
• support for remote indexes
• automatic full-dataset testing each week (#337)
• Improved documentation

Bugfixes

• Supports ziplib on windows (fixes #306)
• Fixed download_info printing issue (fixes #333)
• RWC metadata (#338)
• validation for datasets without metadata (#342)

0.3.0b0

Version 0.3 beta release

• Changes top level API - datasets are now loaded with mirdata.dataset('orchset')
• Adds a Dataset object
• Renames load --> load_tracks
• Adds dataset.choice_track()
• Moves Track and MultiTrack base classes from mirdata.track to mirdata.core
• removes all DATASET_DIR variables and sets default download folder to the module name
• Track object no longer accept data_home=None - defaults are set in the Dataset object
• Fixes bug in beatport_key (#311 )

0.2.2

• fixes bug in orchset download
• adds loader for Mridangam Stroke dataset

0.2.1

Adds a new MultiTrack base class

0.2.0

Compared to the 0.2.0.beta release, 0.2.0 includes:

• A PR template for new loaders
• bugfixes in jams converters
• exhaustive dataset tests
• descriptive assert messages in tests
• a generalized download_utils.downloader() function
• New dataset loaders:
  • groove MIDI
  • Maestro
  • GiantSteps EDM key
  • GiantSteps EDM tempo
  • Beatport EDM key

0.2.0b0

This is the first beta release of the v0.2.X series.

Compared to 0.2.0a0 (alpha release), mirdata 0.2.0b0 has:

• a new utility function jams_utils.notes_to_jams, which converts notes annotations into JAMS format using note_to_midi from librosa. PR #186 by @magdalenafuentes
• better support for remote data. PR #188 by @rabitt
• addition of pysoundfile to the list of requirements. PRs #190 and #193 by @lostanlen
• improvements to the documentation. PRs #194 and #201 by @rabitt
• addition of a pull request template and updates to the contribution policy. PR #198 and #199 by @magdalenafuentes
• a table of all supported datasets. PR #203 by @rabitt
• a base Track class with unified __repr__ implementation. PR #205 by @rabitt

To consult the list of supported datasets, please visit:
https://mirdata.readthedocs.io/en/latest/source/datasets.html

To contribute a new dataset to mirdata, please visit:
https://github.com/mir-dataset-loaders/mirdata/blob/master/CONTRIBUTING.md

0.2.0a0

This is the first alpha release of the v0.2.X series.

Compared to v0.1, mirdata v0.2.0a0 has a new loader: TinySOL (#165). TinySOL is a dataset of 2478 samples, each containing a single musical note from one of 14 different instruments.

Furthermore, @rabitt has added a new data type in mirdata called utils.MultipitchData, which is compatible with mir_eval's multipitch.evaluate function. This addition fixes issue #101: "medleydb_melody melody3 type not in mir_eval format".

Lastly, @magdalenafuentes made improvements to the documentation.

v0.1.2

This version fixes a major bug in v0.1.0, related to the index of the RWC-Genre dataset.

As noted in issue #80, the file indexes/rwc_genre.json was incomplete: it contained had only 90 items instead of 100. More precisely, it lacked files RM-G09{i} for i in {1,..,9} and RM-G100. If these files were missing from DATA_HOME, or had invalid checksums, the function mirdata.rwc_genre.validate would have returned ({}, {}), i.e. as if all files were present and valid. In other words, the validator of the mirdata.rwc_genre was too lenient. In the interest of reproducible usage, we recommend you to stop relying on the mirdata.rwc_genre module to validate the RWC-genre dataset.

For this reason, we have removed the mirdata.rwc_genre module entirely in v0.1.2.