This repository contains DQN algorithm for pogema. Algorithm uses logger for training on previous experiments and two NNs: target net and policy net. Policy net is being training every training step and once in TARGET_UPDATE
steps is being logged into target net for stable learning. File vis.py
contains script for visualizing results into .svg
file.
supercrablover / dqn_for_pogema Goto Github PK
View Code? Open in Web Editor NEWDeep Q-Learning algorithm for Partially-Observable Grid Environment for Multiple Agents