Weight update step for reward/loss based learning under bandit settings

gistfile1.txt

lossScalar = 1 - reward; % This is loss of the chosen action
lossVector = zeros(1,self.nbActions);
lossVector(astAction) = lossScalar;
self.timeStep=self.timeStep+1;
%The weight update step below depends on the learning policy. This will probably be handled by the NN/RL-net
self.weights=self.weights.*(exp(-sqrt(log(self.numActions)/self.timeStep)*lossVector))';