Check sliding time window length in variation regressors

[jrasero]

No description provided.

[jrasero]

Here, as they were defined t_i and t_f, the window length is of 2*T_R , but according to Chang2009 we should use a 3*T_R window.

https://github.com/CoAxLab/brainhack-physio-project/blob/4de9c91a63ccc43135700dfe713ca5fe2e4623ef/code/regressors.py#L405

https://github.com/CoAxLab/brainhack-physio-project/blob/4de9c91a63ccc43135700dfe713ca5fe2e4623ef/code/regressors.py#L536

[jrasero]

A new argument time_window has been added to the Variation models, with a default value of 6 s (i.e. a 3*T_R window for a common T_R=2 s). What it does is to take that window length, center at each T_R time tick, to compute both the standard deviation in signal and the differences between events inside the window.

In this way, we have some flexibility and could avoid some problematic scenarios that happened due to the time window being defined in terms of the T_R. For example, for fast acquisitions like the human connectome (T_R=0.72 s) the time window was very narrow, so many times, particularly with variations in heart rate, there was be just one or none events in that window, and therefore non regressors were produced for this case.

Curious to know your thoughts on this @timothyv, @coaxlab0