In the IRASA paper, the algorithm section notes that they split the signal into 10 overlapping segments of 90% of the signal, run IRASA on each segment, then average the resulting psd and psd_aperiodic across segments. The current IRASA implementation in neurodsp doesn't segment the signal. This may be worth adding. Kwargs for n_segs and seg_fract could be added so n_segs=1 and seg_fract=1 would produce the same results as the current implementation, while n_segs=10 and seg_fract=.1 would reproduce the paper's algorithm.
In the IRASA paper, the algorithm section notes that they split the signal into 10 overlapping segments of 90% of the signal, run IRASA on each segment, then average the resulting psd and psd_aperiodic across segments. The current IRASA implementation in neurodsp doesn't segment the signal. This may be worth adding. Kwargs for
n_segsandseg_fractcould be added son_segs=1andseg_fract=1would produce the same results as the current implementation, whilen_segs=10andseg_fract=.1would reproduce the paper's algorithm.