I fixed a bug in the track_policy function that would lead the agent to cycle continuously when no close enough vector could be found. Now, a single pass is made over all Q-sets and the closest vector is returned as the new target. I also made a small optimisation that makes the agent break out of the loop when a vector is found that matches close enough, to avoid having to do an exhaustive search when possible.
ffelten
commented
1 year ago
I fixed a bug in the
track_policyfunction that would lead the agent to cycle continuously when no close enough vector could be found. Now, a single pass is made over all Q-sets and the closest vector is returned as the new target. I also made a small optimisation that makes the agent break out of the loop when a vector is found that matches close enough, to avoid having to do an exhaustive search when possible.