Table A1.

List of hyper-parameters.

Parameter	Value
The length of CGM measurements L	6
The hidden units of DNNs	[200, 200, 10]
The learning rate of the actor	0.0001
The learning rate of the critic	0.0001
The size of replay memory N	500
Batch size	32
Soft replacement $\tau$	0.01
Target network update period $\mathcal{T}$	100
Discount factor $\gamma$	0.9
The degree of prioritization $\alpha$	0.6
Compensation factor $\beta$	$0.4\to 1$
Priority constant $ϵ$	0.00001