Contributing

We encourage contributions to mirdata, especially new dataset loaders. To contribute a new loader, follow the steps indicated below and create a Pull Request (PR) to the github repository.

Installing and running tests

First, clone the repository from github:

git clone git@github.com:mir-dataset-loaders/mirdata.git

We recommend you install pyenv to manage your Python versions and install all mirdata requirements. You will want to install the latest versions of Python 3.6 and 3.7. Once pyenv and the Python versions are configured, install pytest. Finally, run :

pytest tests/ --local

All tests should pass!

Writing a new dataset loader

The steps to add a new dataset loader to mirdata are:

  1. Create an index
  2. Create a module
  3. Add tests
  4. Submit your loader

Before starting, if your dataset is not fully downloadable you should:

  1. Contact the mirdata team by opening an issue or PR so we can discuss how to proceed with the closed dataset.
  2. Show that the version used to create the checksum is the “canonical” one, either by getting the version from the dataset creator, or by verifying equivalence with several other copies of the dataset.

To reduce friction, we will make commits on top of contributors PRs by default unless the please-do-not-edit flag is used.

1. Create an index

mirdata’s structure relies on JSON objects called indexes. Indexes contain information about the structure of the dataset which is necessary for the loading and validating functionalities of mirdata. In particular, indexes contain information about the files included in the dataset, their location and checksums. The necessary steps are:

  1. To create an index, first cereate a script in scripts/, as make_dataset_index.py, which generates an index file.
  2. Then run the script on the canonical versions of the dataset and save the index in mirdata/datasets/indexes/ as dataset_index.json.

Here there is an example of an index to use as guideline:

Example Make Index Script

# -*- coding: utf-8 -*-
import argparse
import glob
import json
import os
from mirdata.validate import md5

DATASET_INDEX_PATH = "../mirdata/datasets/indexes/dataset_index.json"


def make_dataset_index(dataset_data_path):
    annotation_dir = os.path.join(dataset_data_path, "annotation")
    annotation_files = glob.glob(os.path.join(annotation_dir, "*.lab"))
    track_ids = sorted([os.path.basename(f).split(".")[0] for f in annotation_files])

    # top-key level metadata
    metadata_checksum = md5(os.path.join(dataset_data_path, "id_mapping.txt"))
    index_metadata = {"metadata": {"id_mapping": ("id_mapping.txt", metadata_checksum)}}

    # top-key level tracks
    index_tracks = {}
    for track_id in track_ids:
        audio_checksum = md5(
            os.path.join(dataset_data_path, "Wavfile/{}.wav".format(track_id))
        )
        annotation_checksum = md5(
            os.path.join(dataset_data_path, "annotation/{}.lab".format(track_id))
        )

        index_tracks[track_id] = {
            "audio": ("Wavfile/{}.wav".format(track_id), audio_checksum),
            "annotation": ("annotation/{}.lab".format(track_id), annotation_checksum),
        }

    # top-key level version
    dataset_index = {"version": None}

    # combine all in dataset index
    dataset_index.update(index_metadata)
    dataset_index.update({"tracks": index_tracks})

    with open(DATASET_INDEX_PATH, "w") as fhandle:
        json.dump(dataset_index, fhandle, indent=2)


def main(args):
    make_dataset_index(args.dataset_data_path)


if __name__ == "__main__":
    PARSER = argparse.ArgumentParser(description="Make dataset index file.")
    PARSER.add_argument(
        "dataset_data_path", type=str, help="Path to dataset data folder."
    )

    main(PARSER.parse_args())

More examples of scripts used to create dataset indexes can be found in the scripts folder.

tracks

Most MIR datasets are organized as a collection of tracks and annotations. In such case, the index should make use of the tracks top-level key. A dictionary should be stored under the tracks top-level key where the keys are the unique track ids of the dataset. The values are a dictionary of files associated with a track id, along with their checksums. These files can be for instance audio files or annotations related to the track id. File paths are relative to the top level directory of a dataset.

Index Examples - Tracks

If the version 1.0 of a given dataset has the structure:

> Example_Dataset/
    > audio/
        track1.wav
        track2.wav
        track3.wav
    > annotations/
        track1.csv
        Track2.csv
        track3.csv
    > metadata/
        metadata_file.csv

The top level directory is Example_Dataset and the relative path for track1.wav would be audio/track1.wav. Any unavailable fields are indicated with null. A possible index file for this example would be:

{   "version": "1.0",
    "tracks":
        "track1": {
            "audio": [
                "audio/track1.wav",  // the relative path for track1's audio file
                "912ec803b2ce49e4a541068d495ab570"  // track1.wav's md5 checksum
            ],
            "annotation": [
                "annotations/track1.csv",  // the relative path for track1's annotation
                "2cf33591c3b28b382668952e236cccd5"  // track1.csv's md5 checksum
            ]
        },
        "track2": {
            "audio": [
                "audio/track2.wav",
                "65d671ec9787b32cfb7e33188be32ff7"
            ],
            "annotation": [
                "annotations/Track2.csv",
                "e1964798cfe86e914af895f8d0291812"
            ]
        },
        "track3": {
            "audio": [
                "audio/track3.wav",
                "60edeb51dc4041c47c031c4bfb456b76"
            ],
            "annotation": [
                "annotations/track3.csv",
                "06cb006cc7b61de6be6361ff904654b3"
            ]
        },
    }
"metadata": {
        "metadata_file": [
            "metadata/metadata_file.csv",
            "7a41b280c7b74e2ddac5184708f9525b"
        ]
}
}

Note

In this example there is a (purposeful) mismatch between the name of the audio file track2.wav and its corresponding annotation file, Track2.csv, compared with the other pairs. This mismatch should be included in the index. This type of slight difference in filenames happens often in publicly available datasets, making pairing audio and annotation files more difficult. We use a fixed, version-controlled index to account for this kind of mismatch, rather than relying on string parsing on load.

multitracks

Index Examples - Multitracks

Coming soon

records

Index Examples - Records

Coming soon

2. Create a module

Once the index is created you can create the loader. For that, we suggest you use the following template and adjust it for your dataset. To quickstart a new module:

  1. Copy the example below and save it to mirdata/datasets/<your_dataset_name>.py
  2. Find & Replace Example with the <your_dataset_name>.
  3. Remove any lines beginning with # – which are there as guidelines.

Example Module

# -*- coding: utf-8 -*-
"""Example Dataset Loader

.. admonition:: Dataset Info
    :class: dropdown

    Please include the following information at the top level docstring for the dataset's module `dataset.py`:

    1. Describe annotations included in the dataset
    2. Indicate the size of the datasets (e.g. number files and duration, hours)
    3. Mention the origin of the dataset (e.g. creator, institution)
    4. Describe the type of music included in the dataset
    5. Indicate any relevant papers related to the dataset
    6. Include a description about how the data can be accessed and the license it uses (if applicable)

"""

import logging
import os
import json
import librosa
import csv
import numpy as np
# -- import whatever you need here and remove
# -- example imports you won't use

from mirdata import download_utils
from mirdata import jams_utils
from mirdata import core, annotations


# -- Add any relevant citations here
BIBTEX = """@article{article-minimal,
    author = "L[eslie] B. Lamport",
    title = "The Gnats and Gnus Document Preparation System",
    journal = "G-Animal's Journal",
    year = "1986"
}"""

# -- REMOTES is a dictionary containing all files that need to be downloaded.
# -- The keys should be descriptive (e.g. 'annotations', 'audio').
# -- When having data that can be partially downloaded, remember to set up
# -- correctly destination_dir to download the files following the correct structure.
REMOTES = {
    'remote_data': download_utils.RemoteFileMetadata(
        filename='a_zip_file.zip',
        url='http://website/hosting/the/zipfile.zip',
        checksum='00000000000000000000000000000000',  # -- the md5 checksum
        destination_dir='path/to/unzip' # -- relative path for where to unzip the data, or None
    ),
}

# -- Include any information that should be printed when downloading
# -- remove this variable if you don't need to print anything during download
DOWNLOAD_INFO = """
Include any information you want to be printed when dataset.download() is called.
These can be instructions for how to download the dataset (e.g. request access on zenodo),
caveats about the download, etc
"""

# -- Include the dataset's license information
LICENSE_INFO = """
The dataset's license information goes here.
"""

# -- change this to load any top-level metadata
## delete this function if you don't have global metadata
def _load_metadata(data_home):
    metadata_path = os.path.join(data_home, 'example_metadta.csv')
    if not os.path.exists(metadata_path):
        logging.info('Metadata file {} not found.'.format(metadata_path))
        return None

    # load metadata however makes sense for your dataset
    metadata_path = os.path.join(data_home, 'example_metadata.json')
    with open(metadata_path, 'r') as fhandle:
        metadata = json.load(fhandle)

    metadata['data_home'] = data_home

    return metadata


DATA = core.LargeData('example_index.json', _load_metadata)
# DATA = core.LargeData('example_index.json')  ## use this if your dataset has no metadata


class Track(core.Track):
    """Example track class
    # -- YOU CAN AUTOMATICALLY GENERATE THIS DOCSTRING BY CALLING THE SCRIPT:
    # -- `scripts/print_track_docstring.py my_dataset`
    # -- note that you'll first need to have a test track (see "Adding tests to your dataset" below)

    Args:
        track_id (str): track id of the track

    Attributes:
        track_id (str): track id
        # -- Add any of the dataset specific attributes here

    """
    def __init__(self, track_id, data_home):
        if track_id not in DATA.index:
            raise ValueError(
                '{} is not a valid track ID in Example'.format(track_id))

        self.track_id = track_id

        self._data_home = data_home
        self._track_paths = DATA.index[track_id]

        # -- add any dataset specific attributes here
        self.audio_path = os.path.join(
            self._data_home, self._track_paths['audio'][0])
        self.annotation_path = os.path.join(
            self._data_home, self._track_paths['annotation'][0])

        # -- if the user doesn't have a metadata file, load None
        self._metadata = DATA.metadata(data_home)
        if self._metadata is not None and track_id in self._metadata:
            self.some_metadata = self._metadata[track_id]['some_metadata']
        else:
            self.some_metadata = None

    # -- `annotation` will behave like an attribute, but it will only be loaded
    # -- and saved when someone accesses it. Useful when loading slightly
    # -- bigger files or for bigger datasets. By default, we make any time
    # -- series data loaded from a file a cached property
    @core.cached_property
    def annotation(self):
        """output type: description of output"""
        return load_annotation(self.annotation_path)

    # -- `audio` will behave like an attribute, but it will only be loaded
    # -- when someone accesses it and it won't be stored. By default, we make
    # -- any memory heavy information (like audio) properties
    @property
    def audio(self):
        """(np.ndarray, float): DESCRIPTION audio signal, sample rate"""
        return load_audio(self.audio_path)

    # -- we use the to_jams function to convert all the annotations in the JAMS format.
    # -- The converter takes as input all the annotations in the proper format (e.g. beats
    # -- will be fed as beat_data=[(self.beats, None)], see jams_utils), and returns a jams
    # -- object with the annotations.
    def to_jams(self):
        """Jams: the track's data in jams format"""
        return jams_utils.jams_converter(
            audio_path=self.audio_path,
            annotation_data=[(self.annotation, None)],
            metadata=self._metadata,
        )
        # -- see the documentation for `jams_utils.jams_converter for all fields


# -- if the dataset contains multitracks, you can define a MultiTrack similar to a Track
# -- you can delete the block of code below if the dataset has no multitracks
class MultiTrack(core.MultiTrack):
    """Example multitrack class

    Args:
        mtrack_id (str): multitrack id
        data_home (str): Local path where the dataset is stored.
            If `None`, looks for the data in the default directory, `~/mir_datasets/Example`

    Attributes:
        mtrack_id (str): track id
        tracks (dict): {track_id: Track}
        track_audio_attribute (str): the name of the attribute of Track which
            returns the audio to be mixed
        # -- Add any of the dataset specific attributes here

    """
    def __init__(self, mtrack_id, data_home):
        self.mtrack_id = mtrack_id
        self._data_home = data_home
        # these three attributes below must have exactly these names
        self.track_ids = [...] # define which track_ids should be part of the multitrack
        self.tracks = {t: Track(t, self._data_home) for t in self.track_ids}
        self.track_audio_property = "audio" # the property of Track which returns the relevant audio file for mixing

        # -- optionally add any multitrack specific attributes here
        self.mix_path = ...  # this can be called whatever makes sense for the datasets
        self.annotation_path = ...

    # -- multitracks can optionally have mix-level cached properties and properties
    @core.cached_property
    def annotation(self):
        """output type: description of output"""
        return load_annotation(self.annotation_path)

    @property
    def audio(self):
        """(np.ndarray, float): DESCRIPTION audio signal, sample rate"""
        return load_audio(self.audio_path)

    # -- multitrack objects are themselves Tracks, and also need a to_jams method
    # -- for any mixture-level annotations
    def to_jams(self):
        """Jams: the track's data in jams format"""
        return jams_utils.jams_converter(
            audio_path=self.mix_path,
            annotation_data=[(self.annotation, None)],
            ...
        )
        # -- see the documentation for `jams_utils.jams_converter for all fields


def load_audio(audio_path):
    """Load a Example audio file.

    Args:
        audio_path (str): path to audio file

    Returns:
        * np.ndarray - the mono audio signal
        * float - The sample rate of the audio file

    """
    # -- for example, the code below. This should be dataset specific!
    # -- By default we load to mono
    # -- change this if it doesn't make sense for your dataset.
    if not os.path.exists(audio_path):
        raise IOError("audio_path {} does not exist".format(audio_path))
    return librosa.load(audio_path, sr=None, mono=True)


# -- Write any necessary loader functions for loading the dataset's data
def load_annotation(annotation_path):

    # -- if there are some file paths for this annotation type in this dataset's
    # -- index that are None/null, uncomment the lines below.
    # if annotation_path is None:
    #     return None

    if not os.path.exists(annotation_path):
        raise IOError("annotation_path {} does not exist".format(annotation_path))

    with open(annotation_path, 'r') as fhandle:
        reader = csv.reader(fhandle, delimiter=' ')
        intervals = []
        annotation = []
        for line in reader:
            intervals.append([float(line[0]), float(line[1])])
            annotation.append(line[2])

    annotation_data = annotations.EventData(
        np.array(intervals), np.array(annotation)
    )
    return annotation_data

# -- use this decorator so the docs are complete
@core.docstring_inherit(core.Dataset)
class Dataset(core.Dataset):
    """The Example dataset
    """

    def __init__(self, data_home=None):
        super().__init__(
            data_home,
            index=DATA.index,
            name="Example",
            track_object=Track,
            bibtex=BIBTEX,
            remotes=REMOTES,
            download_info=DOWNLOAD_INFO,
        )

    # -- Copy any loaders you wrote that should be part of the Dataset object
    # -- use this core.copy_docs decorator to copy the docs from the original
    # -- load_ function
    @core.copy_docs(load_audio)
    def load_audio(self, *args, **kwargs):
        return load_audio(*args, **kwargs)

    @core.copy_docs(load_annotation)
    def load_annotation(self, *args, **kwargs):
        return load_annotation(*args, **kwargs)

# -- if your dataset needs to overwrite the default download logic, do it here.
# -- this function is usually not necessary unless you need very custom download logic
def download(
    self, partial_download=None, force_overwrite=False, cleanup=False
):
    """Download the dataset

    Args:
        partial_download (list or None):
            A list of keys of remotes to partially download.
            If None, all data is downloaded
        force_overwrite (bool):
            If True, existing files are overwritten by the downloaded files. 
        cleanup (bool):
            Whether to delete any zip/tar files after extracting.

    Raises:
        ValueError: if invalid keys are passed to partial_download
        IOError: if a downloaded file's checksum is different from expected

    """
    # see download_utils.downloader for basic usage - if you only need to call downloader
    # once, you do not need this function at all.
    # only write a custom function if you need it!

You may find these examples useful as references:

For many more examples, see the datasets folder.

3. Add tests

To finish your contribution, include tests that check the integrity of your loader. For this, follow these steps:

  1. Make a toy version of the dataset in the tests folder tests/resources/mir_datasets/my_dataset/, so you can test against little data. For example:

    • Include all audio and annotation files for one track of the dataset
    • For each audio/annotation file, reduce the audio length to 1-2 seconds and remove all but a few of the annotations.
    • If the dataset has a metadata file, reduce the length to a few lines.

  2. Test all of the dataset specific code, e.g. the public attributes of the Track object, the load functions and any other custom functions you wrote. See the tests folder for reference. If your loader has a custom download function, add tests similar to this loader.

  3. Locally run pytest -s tests/test_full_dataset.py --local --dataset my_dataset before submitting your loader to make sure everything is working.

Note

We have written automated tests for all loader’s cite, download, validate, load, track_ids functions, as well as some basic edge cases of the Track object, so you don’t need to write tests for these!

Running your tests locally

Before creating a PR, you should run all the tests locally like this:

pytest tests/ --local

The –local flag skips tests that are built to run only on the remote testing environment.

To run one specific test file:

pytest tests/test_ikala.py

Finally, there is one local test you should run, which we can’t easily run in our testing environment.

pytest -s tests/test_full_dataset.py --local --dataset dataset

Where dataset is the name of the module of the dataset you added. The -s tells pytest not to skip print statments, which is useful here for seeing the download progress bar when testing the download function.

This tests that your dataset downloads, validates, and loads properly for every track. This test takes a long time for some datasets, but it’s important to ensure the integrity of the library.

We’ve added one extra convenience flag for this test, for getting the tests running when the download is very slow:

pytest -s tests/test_full_dataset.py --local --dataset my_dataset --skip-download

which will skip the downloading step. Note that this is just for convenience during debugging - the tests should eventually all pass without this flag.

Working with big datasets

In the development of large datasets, it is advisable to create an index as small as possible to optimize the implementation process of the dataset loader and pass the tests.

Working with remote indexes

For the end-user there is no difference between the remote and local indexes. However, indexes can get large when there are a lot of tracks in the dataset. In these cases, storing and accessing an index remotely can be convenient.

However, to contribute to the library using remote indexes you have to add in utils.LargeData(...) the remote_index argument with a download_utils.RemoteFileMetadata dictionary with the remote index information.

REMOTE_INDEX = {
    "REMOTE_INDEX": download_utils.RemoteFileMetadata(
        filename="acousticbrainz_genre_index.json.zip",
        url="https://zenodo.org/record/4298580/files/acousticbrainz_genre_index.json.zip?download=1",
        checksum="810f1c003f53cbe58002ba96e6d4d138",
        destination_dir="",
    )
}
DATA = utils.LargeData("acousticbrainz_genre_index.json", remote_index=REMOTE_INDEX)

Reducing the testing space usage

We are trying to keep the test resources folder size as small as possible, because it can get really heavy as new loaders are added. We kindly ask the contributors to reduce the size of the testing data if possible (e.g. trimming the audio tracks, keeping just two rows for csv files).

4. Submit your loader

Before you submit your loader make sure to:

  1. Add your module to docs/source/mirdata.rst following an alphabetical order
  2. Add your module to docs/source/table.rst following an alphabetical order as follows:
* - Dataset
  - Downloadable?
  - Annotation Types
  - Tracks
  - License

An example of this for the Beatport EDM key dataset:

* - Beatport EDM key
  - - audio: ✅
    - annotations: ✅
  - - global :ref:`key`
  - 1486
  - .. image:: https://licensebuttons.net/l/by-sa/3.0/88x31.png
       :target: https://creativecommons.org/licenses/by-sa/4.0

(you can check that this was done correctly by clicking on the readthedocs check when you open a PR). You can find license badges images and links here.

Pull Request template

When starting your PR please use the new_loader.md template, it will simplify the reviewing process and also help you make a complete PR. You can do that by adding &template=new_loader.md at the end of the url when you are creating the PR :

...mir-dataset-loaders/mirdata/compare?expand=1 will become ...mir-dataset-loaders/mirdata/compare?expand=1&template=new_loader.md.

Docs

Staged docs for every new PR are built, and you can look at them by clicking on the “readthedocs” test in a PR. To quickly troubleshoot any issues, you can build the docs locally by nagivating to the docs folder, and running make html (note, you must have sphinx installed). Then open the generated _build/source/index.html file in your web browser to view.

Troubleshooting

If github shows a red X next to your latest commit, it means one of our checks is not passing. This could mean:

  1. running black has failed – this means that your code is not formatted according to black’s code-style. To fix this, simply run the following from inside the top level folder of the repository:
black --target-version py37 --skip-string-normalization mirdata/
  1. the test coverage is too low – this means that there are too many new lines of code introduced that are not tested.
  2. the docs build has failed – this means that one of the changes you made to the documentation has caused the build to fail. Check the formatting in your changes and make sure they are consistent.
  3. the tests have failed – this means at least one of the tests is failing. Run the tests locally to make sure they are passing. If they are passing locally but failing in the check, open an issue and we can help debug.

Documentation

This documentation is in rst format. It is built using Sphinx and hosted on readthedocs. The API documentation is built using autodoc, which autogenerates documentation from the code’s docstrings. We use the napoleon plugin for building docs in Google docstring style. See the next section for docstring conventions.

mirdata uses Google’s Docstring formatting style. Here are some common examples.

Note

The small formatting details in these examples are important. Differences in new lines, indentation, and spacing make a difference in how the documentation is rendered. For example writing Returns: will render correctly, but Returns or Returns : will not.

Functions:

def add_to_list(list_of_numbers, scalar):
    """Add a scalar to every element of a list.
    You can write a continuation of the function description here on the next line.

    You can optionally write more about the function here. If you want to add an example
    of how this function can be used, you can do it like below.

    Example:
        .. code-block:: python

        foo = add_to_list([1, 2, 3], 2)

    Args:
        list_of_numbers (list): A short description that fits on one line.
        scalar (float):
            Description of the second parameter. If there is a lot to say you can
            overflow to a second line.

    Returns:
        list: Description of the return. The type here is not in parentheses

    """
    return [x + scalar for x in list_of_numbers]

Functions with more than one return value:

def multiple_returns():
    """This function has no arguments, but more than one return value. Autodoc with napoleon doesn't handle this well,
    and we use this formatting as a workaround.

    Returns:
        * int - the first return value
        * bool - the second return value

    """
    return 42, True

One-line docstrings

def some_function():
    """
    One line docstrings must be on their own separate line, or autodoc does not build them properly
    """
    ...

Objects

"""Description of the class
overflowing to a second line if it's long

Some more details here

Args:
    foo (str): First argument to the __init__ method
    bar (int): Second argument to the __init__ method

Attributes:
    foobar (str): First track attribute
    barfoo (bool): Second track attribute

Cached Properties:
    foofoo (list): Cached properties are special mirdata attributes
    barbar (None): They are lazy loaded properties.
    barf (bool): Document them with this special header.

"""

Conventions

We use the following libraries for loading data from files:

Format library
audio (wav, mp3, …) librosa
midi pretty_midi
json json
csv csv
jams jams

Custom track attributes should be global, track-level data. For some datasets, there is a separate, dataset-level metadata file with track-level metadata, e.g. as a csv. When a single file is needed for more than one track, we recommend using writing a _load_metadata method and passing it to a LargeData object, which is available globally throughout the module to avoid loading the same file multiple times.

Track properties and cached properties should be trivial, and directly call a load_* method. There should be no additional logic in a track property/cached property, and instead all logic should be done in the load method. We separate these because the track properties are only usable when data is available locally - when data is remote, the load methods are used instead.

If a Track has a property, for example a type of annotation, that is present for some tracks and not others, the property should be set to None when it isn’t available.

The index should only contain key-values for files that exist.

Custom Decorators

This is used primarily for Track objects.

This decorator causes an Object’s function to behave like an attribute (aka, like the @property decorator), but caches the value in memory after it is first accessed. This is used for data which is relatively large and loaded from files.

This decorator is used for children of the Dataset object, and copies the Attributes from the parent class to the docstring of the child. This gives us clear and complete docs without a lot of copy-paste.

This decorator is used mainly for a dataset’s load_ functions, which are attached to a loader’s Dataset object. The attached function is identical, and this decorator simply copies the docstring from another function.