neuromore/studio

Latency Analysis

Opened this issue · 0 comments

Hi,

Has anyone ever undertaken an analysis of the latency of the Studio?

I am about to look at the data/runtime in more detail, and see if I can find the roundtrip from data coming off the board to the point my corresponding feedback score gets output via OSC, but personally I'm starting to find some evidence that low-latency classifiers may work better for training (no epoching or smoothing etc. - as much as possible the feedback score right now is the current bandpower adjusted by the mean and standard deviation to fit within [-1,+1]).

Still not fully sure how to interpret the information in the plots, but I have some evidence from the BDF data output that it may take about 0.5s from bandpower to feedback for my classifier, which is gonna invalidate this line of enquiry. I think it'd be good to see how much the CPU is spending in various functions, and also if I can follow a timestamp through the graph and time when it exits.

Is this something that's ever been looked at within the Engine at all?

Cheers