# Multi-armed Bandit Algorithm selection and Optimization

I have 2 channels that I can sent my products, the A channel cost 0.10\$per product and the B costs 0.01\$ and I am trying dynamically to optimize the channel selection by minimize the cost.

Obviously the B channel is a proper selection as the cost is lower but there is some issues with this channel. As the B channel is not a direct channel there is a probability to lose the product and we wont get paid if the customer does not get the product and as a result we lose the profit, cost of the product and the cost to ship the product though this channel.

The optimum results from my algorithm I want to be like sent 30% from your products through the A channel and 70% through the B channel and I want to change through time.

A Multi-armed Bandit Algorithm is proper to dynamically change channels over time?

If Bandit is the answer which implementation do you propose (epsilon-degree or Thompson sampling) ?

What do you suggest in order to accomplish that ?

• Can you specify your utility function and all your constraints? – ExabytE Jun 11 '19 at 14:49
• I'm 1 step before that so I don't have yet these things. But I think you can answer that with situation I describe. – dapo Jun 18 '19 at 8:53