Take a look at our :ref:`contributing` docs!
We don't distribute data ourselves, so unfortunately it is up to you to find the data yourself. We strongly encourage you to favor datasets which are currently available.
No, we do not host or distribute datasets.
Open an issue and tag it with the "New Loader" label.
Very often, data fails vaildation because of how the files are named or how the folder is structured. If this is the case, try renaming/reorganizing your data to match what mirdata expects. If your data fails validation because of the checksums, this means that you are using data which is different from what most people are using, and you should try to get the more common dataset version, for example by using the data loader's download function.
Whenever possible, the data downloaded using :code:.download()
is the same data used to create the checksums. If this isn't possible, we did our best to get the data from the original source (the dataset creator) in order to create the checksum. If this is again not possible, we found as many versions of the data as we could from different users of the dataset, computed checksums on all of them and used the version which was the most common amongst them.
For now, no. Music datasets are very widely varied in their annotation types and supported tasks. To make a data loader, there would need to be "standard" ways to encode the desired inputs/outputs - unfortunately this is not universal for most datasets and usages. Still, this library provides the necessary first step for building data loaders and it is easy to build data loaders on top of this. For more information, see :ref:`Using mirdata with tensorflow`.
Please open an issue and tag it with the "broken link" label.
mirdata = mir + data. MIR is an acronym for Music Information Retrieval, and the library was built for working with data.
No. All datasets have "mistakes", and we do not want to create another version of each dataset ourselves. The loaders should load the data as released. After that, it's up to the user what they want to do with it.
Yes! We use the smart_open library, which supports non-local filesystems such as GCS and AWS. See :ref:`Remote Data Example` for details.