Table 1.

Comparative overview for all detection/estimation analysis procedures available in pyhrf in terms of model structure and analysis duration.

	BS GLM	FIR GLM	RFIR	JDE
Spatial regularization	Smoothing	Smoothing	Smoothing	Adaptive
Temporal regularization	None	None	2nd order deriv.	2nd order deriv.
HRF shape constraint	Function basis	Free	Smooth	Smooth
Number of unknowns for the stimulus-induced component	J × R × M	J × D × M	J × M × (D + 1)	2 × J × M + Γ × (D + 4M + 1)
	1 ⩽ R ⩽ 3	D ≈ 10	10 ⩽ D ⩽ 50	10 ⩽ D ⩽ 50, Γ ≈ 400
Typical ratio of unknowns/data	0.23	0.78	3.4	0.16
Analysis duration	3 min	5 min	30 h	8 h

“2nd order deriv.” stands for a penalization on the energy of the HRF which penalizes abrupt shape changes. The number of nuisance parameters was considered the same for all models, so that only the modeling of the stimulus-induced component is relevant to assess model parsimony. The ratio “unknowns/data” is given for a typical fMRI data analysis with J = 4 × 10⁴, R = 3, D = 40, M = 10, Γ = 400, and N = 128 (total number of data points: N × J). The analysis duration is for a whole brain data treatment on an Intel Core i5 (M480 2.67 Ghz).