ENH: Make dataset description

[effigies]

Following BEP 003:

Derived dataset and pipeline description

<dataset>/derivatives/<pipeline_name>/dataset_description.json

Keys:

• HowToAcknowledge
• PipelineDescription
  • Name
  • Version
  • CodeURL
  • DockerHubContainerTag
  • SingularityContainerURL
  • SingularityContainerMD5
• SourceDatasetsURLs - a list of URLs to the source dataset(s) (mandatory)
• SourceDatasetsVersions - a list of versions of the source dataset(s) (optional only if SourceDatasetsURLs specify the version unequivocally)
• License

We should start adding docker hub tags and singularity URLs as environment variables when we build images. I've put in a stub to get the MD5, which will attempt to pull the 'version' tag from Singularity Hub.

BIDS doesn't specify a field for version, which seems like an oversight, but perhaps there's another mechanism for marking it?

Long-term, it would be good to standardize the construction of this metadata in the BIDS-App spec and pybids.

Related to discussion in #28.

[adelavega]

Okay, I think you're right about this. I was mostly confused about the semantics of calling this a "dataset", but I see that such a file was recommended in the original derivatives BEP. Otherwise, it very much makes sense to include such meta-data in the outputs.

[adelavega]

Looks good

[adelavega]

You can omit this line, since the dir is also made in write_derivative_description

[effigies]

Does it make more sense for write_derivative_description to assume that deriv_dir exists?

[adelavega]

Probably, although that's what's nice about exist_ok, is you can use it either way.

[adelavega]

If you think in the future there may be other files you might want to instantiate in the folder, you could call the function something like create_derivative_template or something

[effigies]

Okay. For now I'm just going to let create_workflow do the makedirs. Might change again in the future.