This epsilon-greedy [1] thing looks just like what my old boss use to tell me, 'trust but verify'. ohh.. I got so sick of those words, but at least now I have an algorithm for it.
Just change the epsilon = 0.1 (10%) higher or lower depending on your initial (personal) confidence, and if your guess was right, and your epsilon low, then the overall impact to 'optimal' solution is negligible, but you have built in a fail safe in case you were human after all.
Just change the epsilon = 0.1 (10%) higher or lower depending on your initial (personal) confidence, and if your guess was right, and your epsilon low, then the overall impact to 'optimal' solution is negligible, but you have built in a fail safe in case you were human after all.
[1] https://en.wikipedia.org/wiki/Multi-armed_bandit