Best way to store arbitrary table-like data

[jonahpearl]

hi all — I’m wondering about the best way to store table-like (column-based) behavioral data.
From what I can tell, the options are:

1. make a BehavioralTimeSeries and store a bunch of TimeSeries in it, with their timestamps linked to save space and to indicate that they share a common time base;

or

2. make a dynamic table directly from hdmf and store that in the nwb. Is there a more high-level way to use a tabular format for behavioral data? If so would it provide any advantage?

For example, I might want to store a table of timestamped sensor data; or a table with information about each frame of a video. Thanks 🙂

[jonahpearl]

Answer from Ben D:

If your data is trials-like, you should use the trials table. The trials table has a type TimeIntervals, which is a sub-class of DynamicTable. TimeIntervals tables require you to input start_time and stop_time for each row, and you can add any other custom columns you want to each row, including floats, ints, strings, or arrays of those types.
If your data is intervals, but not trials-like (e.g. frames of a video), or if you have a need for multiple tables, you can create multiple TimeIntervals objects and add them to NWBFile.intervals.