Table 1.

Results of training the bitkey data split with different annealing schedules.

Annealing schedule†	Stop‡	Train§			Test#			Validate¶			Average epochs††
		r2‡‡	MAE§§	SE##	r2	MAE	SE	r2	MAE	SE
ON 0.25	T	0.96	30	38	0.89	51	61	0.86	56	67	585
O N 0 .1	T	0.96	29	37	0.89	50	60	0.87	54	65	1018
ON 0.01	T	0.96	28	35	0.89	48	59	0.87	52	66	7375
ON 0.25, BT 0.1, ON 0.01	T	0.98	20	27	0.9	46	57	0.87	52	65	1685
ON 0.25	V	0.96	30	38	0.86	54	66	0.89	51	61	543
O N 0 .1	V	0.96	29	37	0.87	54	65	0.88	51	61	920
ON 0.01	V	0.96	28	36	0.86	54	67	0.89	49	60	7065
ON 0.25, BT 0.1, ON 0.01	V	0.97	26	33	0.87	53	66	0.90	46	59	1313

^†Schedule of changes in learning rate and weight update method as training progresses.

^‡Statistical set used to determine when to change learning rate and weight update method.

^§Training set statistics.

^#Test set statistics.

^¶Validation set statistics.

^††Average of epochs trained over all 40 models.

^‡‡Square of Pearson’s correlation coefficient.

^§§Mean absolute error.

^##Standard error (root mean square error).