koulanurag / minimal-marl

Minimal implementation of multi-agent reinforcement learning algorithms
MIT License
50 stars 11 forks source link

Doubt in train step #5

Open johnMinelli opened 2 years ago

johnMinelli commented 2 years ago

Hi, in this line https://github.com/koulanurag/minimal-marl/blob/bd4ac9360aa686f105d1c3a0f39192ecd824e69d/vdn.py#L101 shouldn't only the active agents contribute to the summation? i.e. (max_q_prime * (1 - done[:, step_i])).sum(dim=1, keepdims=True)