Table 5.

Performance metrics on all assemblies of the standard-discretized dataset, for the studied MOGP models (all with q = 16 latent processes) and three baselines. The best and worst value for each metric are highlighted in green and red respectively. Definitions of the metrics are given in Section 3.2; they are all averaged across all HXS’s. The Err^Σ metrics were not computed for the polynomial method, which was tested at a time when concentration data was unavailable. The training time T_train is defined at the level of one assembly. n is the number of datapoints used for training.

Model	R2	RMSE	Errmax	Err${}_{\mathrm{mean}}^{\mathrm{\Sigma }}$	Err${}_{\max }^{\mathrm{\Sigma }}$	α-CI	PVA	Ttrain (s)	ℛ	n
				(cm−1)	(cm−1)
ICM	0.997	2.01 ⋅ 10−3	6.14 ⋅ 10−2	2.37 ⋅ 10−4	0.363	0.824	−2.15	354	142	201
VLMC	0.994	1.61 ⋅ 10−3	5.10 ⋅ 10−2	1.71 ⋅ 10−4	0.957	1.00	−5.92	260	167	201
PLMC	0.995	1.56 ⋅ 10−3	2.62 ⋅ 10−2	1.52 ⋅ 10−4	0.211	1.00	−7.20	668	141	201
Lazy-LMC	0.999	1.34 ⋅ 10−3	3.23 ⋅ 10−2	8.13 ⋅ 10−5	0.173	1.00	−11.6	5 ⋅ 10−4	142	201
MLI	0.998	2.20 ⋅ 10−3	4.47 ⋅ 10−2	1.69 ⋅ 10−4	9.57 ⋅ 10−2	–	–	0.806	1	4293
BPR	0.998	2.80 ⋅ 10−3	0.102	Not computed	Not computed	0.982	−0.792	3.90	13	201
Best SOGP	0.999	1.36 ⋅ 10−3	3.03 ⋅ 10−2	3.75 ⋅ 10−5	4.81 ⋅ 10−2	0.945	−0.625	16 600	19	201