smarter reading of VectorIndex

[bendichter]

We could use a more convenient way to read the corresponding data for specific row(s) of indexed vectors. We have util.read_indexed_column, which works OK, but it would be better to have this feature as a method of VectorIndex, similar to the refresh and deref methods.

[lawrence-mbf]

It will be difficult to just add methods to VectorIndex as it is a generated class. The one way to get around this would be to add a hardcoded override for metaClass but I'm not a big fan of that. If we implement the dynamic table -> nwb table functions then wouldn't it also solve the issue?

[bendichter]

This would address the particular case when you want to read a portion of an indexed vector. table -> nwb DynamicTable wouldn't address this because that is a write function, not a read function. I don't think nwb DynamicTable -> table would address this either because I was imagining that function would read the entire nwb DynamicTable. If we can't attach methods, then maybe the util function is the best we can do for now.

[lawrence-mbf]

Could be a case for https://github.com/NeurodataWithoutBorders/matnwb/issues/98#issuecomment-467636351 as well.

[bendichter]

Yes, it seems like it may be another good case for it.

[lawrence-mbf]

@bendichter The new getRow function in DynamicTable allows you to grab DynamicTable data given a valid id. Are there any other ways that users index into the table that would be helpful?

[bendichter]

@ln-vidrio yes, by row number is equally common. They are often the same but not always, because users are allowed to set their own ids.

[lawrence-mbf]

Oh, this is a good point. Right now addRow actually requires the ids to be auto-incremented I think. Should the id be required to be set and we just check for duplicates?

[bendichter]

In PyNWB most users do not define the ids themselves but allow them to be populated automatically. There is the option of adding ids manually, which is generally used when a user wants the ids to be consistent with another data source. Would that be possible here as well?

[lawrence-mbf]

I'd have to add that in I think. At that point, is the only requirement that it the id is unique or is that optional as well?

[bendichter]

It needs to be both. I'm not 100% sure that we confirm it is unique if it is input manually on the PyNWB side. Maybe it is checked on write

[oruebel]

I'm not 100% sure that we confirm it is unique

If I remember correctly this can be configured in the PyNWB API, i.e., you can ask it to enforce unique IDs (which I believe some types do) but in general the id's are not checked for uniqueness. In most cases I'd expect the id's to be unique though since folks typically let the API set the id's automatically, in which case PyNWB uses row-index as id.

[lawrence-mbf]

Okay, I can leave regular getRow before to get the row by index, then add a keyword argument allowing for an id to query by id instead, which supercedes whatever the index argument is.