On the identifiability of parameters in reinforcement learning models