MIT Press

Table 2

Hyperparameter tuning for adaptive estimators. We train on Muc7_T A and evaluate on its development set. t denotes the total time for training and prediction on the whole dataset. Best parameters are marked by * and the best scores are highlighted in bold. We report the mean absolute error (MAE), the rooted mean squared error (RMSE), Spearman’s ρ, and the coefficient of determination (R²).

Name	Features	MAE	RMSE	R²	ρ	t
RR(α = 0.5 )	BOW	1.85	2.96	0.47	0.73	0.42
RR(α = 0.5 )	S-BERT	1.92	2.84	0.51	0.79	0.04
RR(α = 1 )	BOW	1.80	2.91	0.49	0.74	0.41
RR(α = 1 ) *	S-BERT	1.89	2.82	0.52	0.79	0.04
GP(kernel=Dot + White)	BOW	1.82	2.93	0.48	0.74	257.67
GP(kernel=Dot + White) *	S-BERT	1.80	2.76	0.54	0.81	14.35
GP(kernel=RBF(1.0)	BOW	5.33	6.71	−1.73	−0.12	300.38
GP(kernel=RBF(1.0)	S-BERT	5.33	6.71	−1.73	−0.12	32.66
GBM	BOW	2.07	3.26	0.36	0.68	0.25
GBM *	S-BERT	1.83	2.83	0.52	0.79	2.98

Name	Features	MAE	RMSE	R²	ρ	t
RR(α = 0.5 )	BOW	1.85	2.96	0.47	0.73	0.42
RR(α = 0.5 )	S-BERT	1.92	2.84	0.51	0.79	0.04
RR(α = 1 )	BOW	1.80	2.91	0.49	0.74	0.41
RR(α = 1 ) *	S-BERT	1.89	2.82	0.52	0.79	0.04
GP(kernel=Dot + White)	BOW	1.82	2.93	0.48	0.74	257.67
GP(kernel=Dot + White) *	S-BERT	1.80	2.76	0.54	0.81	14.35
GP(kernel=RBF(1.0)	BOW	5.33	6.71	−1.73	−0.12	300.38
GP(kernel=RBF(1.0)	S-BERT	5.33	6.71	−1.73	−0.12	32.66
GBM	BOW	2.07	3.26	0.36	0.68	0.25
GBM *	S-BERT	1.83	2.83	0.52	0.79	2.98

View Large

This Feature Is Available To Subscribers Only