# How do I determine the optimal policy in a bandit problem with missing contexts?

Suppose I learn an optimal policy $$\pi(a|c)$$ for a contextual multi-armed bandit problem, where the context $$c$$ is a composite of multiple context variables $$c = c_1, c_2, c_3$$. For example, the context is specified by three Bernoulli variables.

Is there any literature on how to determine the optimal policy in the event where I no longer have access to one of the context variables?

• Hello new user. Maybe explaining how the context variable is lost would help, the question become more clear. – Michael Hearn Nov 27 '19 at 5:21
• So, for example, imagine you have a robot who learns to take actions based on inputs which are three bernoulli variables. We then ast it to select an action but only provide the first two variables. For example, its sense for the third variable breaks. Should we marginalize over the expected reward w.r.t the third input variable? – user31663 Nov 27 '19 at 13:55