interpolation length mismatch

[samdeoxys1]

Is this the expected behavior?

t_l length is 5082, but interpolating head_direction at t_l gives a shorter tsd with 5078

Thanks!

[gviejo]

I think it depend on which time support was used. If you create t_l with the time support of head_direction is it the same?

[samdeoxys1]

Yeah I think the time support might be different. This situation comes up a lot. For instance when the behavior label Tsd have a slightly different time support than some behavior epoch IntervalSet (which happens in many datasets). So I have to manually align things again after doing .interpolate.
I wonder if it would be more convenient to force the result of interpolate to have the same length as the Ts. The points outside of the time support can be made nan.

One additional point: I find myself always calling sth.interpolate(nap.Ts(some_pynapple.t)). It is a bit cumbersome to call nap.Ts all the time. Is it possible to allow the input of interpolate to be np array, and do the conversion inside interpolate?

[wulfdewolf]

Hi @samdeoxys1 , thanks for the message!
Some updates:

• For interpolating, you shouldn't have to align things after using .interpolate(), but if you want the input and output timeseries to have the exact same timepoints you do need to restrict them to the same time support, best to do it before.
  Note: the latest Pynapple release will do the restrict inside the .interpolate() function so you don't have to do it yourself!

• Regarding passing in arrays of timestamps, I understand it's a bit annoying to have to wrap inside a pynapple object, but we do use the time support of that object, so we'll likely keep it that way for now. We'd have to add other arguments to the function otherwise, which we don't want.
  Note: the input does not have to be a Ts only, it can be any Pynapple object, like a TsdFrame, in that case it'll ignore the columns and just use the timestamps!

Let me know if that helps!