I’m running ray.tune with PBT and DQN on a custom environment, which calculates many metrics at the end of each episode. Given that the default Wandb callback plot the metrics at the end of each step for each trial, which consists of many episodes, I was wondering what is a recommended way to plot the metrics at the end of each episode.
More specifically, I have a custom callback based on the Rllib DefaultCallback to calculate the metrics on at the end of each episode. Since all the metrics on a step by step basis in an episode is available there, how can I plot those values, e.g: histograms/plots, on Wandb while running ray tune trials?
Thanks in advance!
P.s: if this question should be moved to Rllib instead, please let me know!