Table 3.

Comparison of the average training results for the str2 and near-backbone-11 alphabets for two architectures specified in Table 1 trained for 50, 150 and 250 epochs

epochs	50	150	250
str2	mean bits saved
IDaaHr	0.84	1.02	1.08
IDGaaH13	0.94	1.10	1.16
near-backbone-11	mean bits saved
IDaaHr	0.05	0.47	0.50
IDGaaH13	0.33	0.51	0.56

The architecture with the guide sequence (IDGaaH13) is consistently superior at all stages of training.