[Question] Custom Eval Callback for train/optimize about rl-baselines3-zoo

[araffin]

Hello,

What is the suggested way for customizing the (Trial)EvalCallback in sb3_zoo?

current recommended way is to fork the RL Zoo, especially if you want to do something really custom.

[kingjin94]

For future reference: I found it was easier to monkey patch the evaluate_policy function for now, i.e., add a own evaluation method to a python configuration file passed via train's conf-file argument and overwrite evaluate_policy, s.a.,

def own_evaluate_policy(model, env, n_eval_episodes, deterministic, render, callback, ...):
    <individual code>

import stable_baselines3.common.callbacks
stable_baselines3.common.callbacks.evaluate_policy = own_evaluate_policy