What is the dimension "num_samples" in ElectricalSeries

[ehennestad]

From the description for ElectricalSeries:

nwb-schema/core/nwb.ecephys.yaml

Lines 4 to 7 in 308c5d8

	doc: A time series of acquired voltage data from extracellular recordings.
	The data field is an int or float array storing data in volts. The first
	dimension should always represent time. The second dimension, if present,
	should represent channels.

It says the data can be 1D or 2D. However, the specification of the data dataset also suggests data can be 3D.

nwb-schema/core/nwb.ecephys.yaml

Lines 17 to 26 in 308c5d8

	datasets:
	- name: data
	dtype: numeric
	dims:
	- - num_times
	- - num_times
	- num_channels
	- - num_times
	- num_channels
	- num_samples

Which is the most up to date? Also what is num_samples in this context? I could understand if it was num_electrodes but num samples is very often the same as num_timepoints so this is confusing (I am not very familiar with ephys, so excuse me if this is obvious)

[CodyCBakerPhD]

I believe the 3D case is for allowing the ability to write only extracted waveform snippets (each of length num_samples, as in number of frames sampled by each waveform) instead of the entire continuous voltage trace (the 2D version)

The most primary use case is the 2D version; there are other more recommended ways of writing waveform data, such as explicitly linking it to spiking events on the units table

In 3.0 I would recommend removing the 1D and 3D cases. I think we also planned at some point last year to add different types for neural ElectricalSeries and non-neural (but still voltage valued) series for things like analog and digital channels

[rly]

@CodyCBakerPhD is correct and I second removing the 1D and 3D cases as well as splitting neural and non-neural electrical time series into different data types.