Predicting Potential Risk: Cerebral Stroke via Regret Minimization

Algorithm 1

Regret minimization experience replay with DQN.

(1)	Initialize feature as .
(2)	Calculate the accuracy of a single feature.
(3)	Initialize replay memory to capacity .
(4)	Initialize action-value function with random weights.
(5)	for episode = 1, do
(6)	Initialize sequence and processed sequenced .
(7)	for do
(8)	Select a feature into as action .
(9)	Execute action in prediction and reward .
(10)	Set and .
(11)	Store transition in .
(12)	Sample random minibatch of transitions
(13)	in .
(14)	end for
(15)	end for