Change default raw BIDS input path from `inputs/data/BIDS` to a BIDS-spec-official location

[asmacdo]

babs currently defaults the raw BIDS input location to inputs/data/BIDS. Neither segment is grounded in the BIDS specification. Two BIDS-study-official locations now exist:

• sourcedata/raw/ — BIDS-study convention for unprocessed raw bids
• rawbids/ — recently added to the BIDS-study to account for the possibility of pre-bids raw data [ENH] Add phenotype and rawbids directories to "study" datasets bids-standard/bids-specification#2191

Proposal: default to one of these. Existing projects with inputs/data/BIDS continue to work via explicit path flag.

[tien-tong]

The default is inputs/data, this is where both the raw BIDS and derivative BIDS go (e.g., fMRIPrep), so with the current setup, people can change it to inputs/data/raw or inputs/data/rawbids.

If we change it rawbids/, where should the input derivative BIDS go?

[asmacdo]

IMO this is a good argument for sourcedata (inputs/data is fine, but going with the standard might make a better default)

so my thinking would be sourcedata/rawbids, sourcedata/somederivative.

@yarikoptic wdyt?

[yarikoptic]

If we change it rawbids/, where should the input derivative BIDS go?

;-) "great minds think alike" ;) -- this was my question/argument "against" rawbids/ as is not "generalizable" to derivatives etc. So you both are right that then it should all just go under sourcedata/. It is primarily when there is a (single?) BIDS raw dataset as the source -- it is allowed (and shown) to have it under rawbids/ so it becomes "clearer". So, overall human (and potentially programs) logic should be

"oh, I see there is rawbids/ -- so this must be a

• "immediate derivative of a BIDS raw dataset since I see sub- subfolders" or
• "a study dataset since I do not see sub- subfolders"

Hiding it under sourcedata/ makes it harder to make such a judgement to decide if it is raw or derived or a study.

Note: In any case, all source datasets should ideally be described (with bids: urls for local ones I believe) in dataset_description.json:SourceDatasets. So, likely that would be the best first step for such discovery before even considering paths based lookup.

[tien-tong]

so my thinking would be sourcedata/rawbids, sourcedata/somederivative.

Sounds good to me!

[yarikoptic]

eh -- rendering broke (I filed issue), IMHO should be rawbids/ at top level

https://bids-specification.readthedocs.io/en/latest/common-principles.html#study-dataset

[asmacdo]

I think the NIDM case might be helpful to consider here, where there are 2 input_datasets.

dataset/
    rawbids/
    sourcedata/NIDM

That feels weird to me, I "like" below better, but I'm open to be convinced.

dataset/
   sourcedata/
       rawbids/
       NIDM/