While training on MAML, I see that the custom metrics from my environment, which I am reporting using the Callback functions, are not been reported to the tensorboard data. The same custom metrics works fine on PPO and other algorithms, but not with MAML.
Also, even though the number of episodes keeps increasing as the training proceed further, the ray_tune logs always report episodes_total: 0. Due to this, the training also doesn’t stop on the stopping criteria which uses episode_total.
Can anyone please help me with how to fix this?