catalystneuro / ndx-ontology-table Goto Github PK

We need a way to uniquely identify elements in the NWBFile so that we can say what the URIs are mapping to within the file. Let's define what features this needs to support.

1. We definitely need to support string types. Are there any other types we need to support? Would it make sense to map a numerical value to a URI? I am leaning towards "no", but I could be convinced otherwise.
2. We definitely need to support datasets, but do we also need to support attributes? I am leaning towards "no."
3. Values in a vector/ regions of a vector. For instance, a Units.cell_type VectorData object might have 30 elements, some of which are "mossy fiber" and some of them "pyramidal" and some "unknown". We should be able to refer to map these specific terms to URIs.

What type of data relationship to we use to identify the fields that are being mapped

1. HDF5 Object Ref ints.
   Cons:

   • These will only be present for HDF5 files, and if we are using another backend they won't work. I'd rather this not be reliant on a specific backend.
   • Attributes are not given Object Ref ints, so this approach would not support attributes.
   • I don't actually know how to expose this int using h5py.

2. Object references/ region references.
   Pros:

   • It seems appropriate to use references for this.
   • automatic validation
   • If we use region references, this supports regions of datasets.

   Cons:

   • I don't think it supports attributes
   • It may be a bit awkward if the references auto-resolve.
   • This is probably going to be difficult in pynwb

3. NWB assigns uuids to each instance of a neurodata_type.
   Pros:

   • Will work the same across different backends
   • Will work for attributes

   Cons:

   • Since the uuid is on the neurodata_type but not its fields, we'll need to add extra steps in order to specify the fields that we are trying to map