ritchiehuang/magail

Pytorch implementation of Multi-Agent Generative Adversarial Imitation Learning

Hi there 👋

I'm currently working as a Maching Learning Engineer intern mainly on reinforcement learning area.

I love reading, writing, watching tv series, especially sleeping 😄.

magail's Issues

Why do we need the variable "masks" in the process of GAE?

It seems that "masks" is used to show whether the reward and the value are at the end of a trajectory. But why do we need it when doing GAE? The code is at the 32nd and 33rd line of GAE.py.
deltas[i, ...] = rewards[i, ...] + gamma * prev_value * masks[i, ...] - values[i, ...]
advantages[i, ...] = deltas[i, ...] + gamma * tau * prev_advantage * masks[i, ...]

>Realization for obtaining expert tracks

“Obtain data from the official project”, can I understand that there is no code realized to obtain expert tracks in the project？

Another question is whether the code implements the decentralized discriminator method?

Thank you for your reply

Originally posted by @yuanyaaa in #3 (comment)

Behavior cloning is very close or better

Hi,

I'm just trying to test MAGAIL as per the paper on MPE env (https://pettingzoo.farama.org/environments/mpe/simple_spread/)
However, when I just train with behavior cloning (supervised learning from states to actions) it trains super fast reaching good accuracy.
When I try with MAGAIL, it gets forever to reach BC level, and sometimes it return to worse performance with further training.

I know this isn't about your code, put perhaps you have an idea of what's going on. I can share my parameters if that will help.

专家轨迹获取

您好！我想请问一下，在MAGAIL中，您是怎么获取专家轨迹的呢？

Custom environment

Thanks for your excellent pytorch implementation of this algorithm.

I have a question if I want to use my own environment. How to remove the environment policy and replace it with my own one?
Do we need to change the optimisation of the policy?

KeyError: 'noise_std'

Hi, it seems like that the hyperparameter noise_std is missing in config.yml.

Could you add it?

Thanks!