We would like to use a Q-learning agent for Pacman (as you did in PA3), but the state space for a large grid is too massive to hold in memory. To solve this, we will switch to feature-based representation of the Pacman game state (similar to PA3 Q10), where we assume that Q(s, a) can be expressed as a (weighted) linear combination of state-action features: Q(s, a)= Σωifi(s, a)

Required:
Suppose we design two features to represent the state (independent of actions): f (s) is the number of ghosts within one step of Pacman, and fp(s) is the number of food pellets within one step of Pacman. Note that we do not have any features dependent on a. Why might this be a bad idea?