Using offline data stored in an .json for evaluation using importance sampling and weighted importance sampling. However, I run into an value error: “ValueError: eps_id 1000 was already passed to the peek function. Make sure the dataset contains only unique episodes with unique ids.”
However, my data set does not have repeated episode IDs beyond those in the same trajectory, for evaluation is it necessary to split these trajectories into single obs, action, and reward episodes? Or is there some other potential issue?