Deep Q-Network (DQN) Learning to Play Atari Space Invaders | Deep Reinforcement Learning | PyTorch
Автор: Raphael Senn
Загружено: 2026-02-06
Просмотров: 103
Описание:
Deep Q-Network (DQN) Learning to Play Atari Space Invaders | Deep Reinforcement Learning | PyTorch
Code: https://github.com/raphaelsenn/atari-dqn
Original paper: https://www.nature.com/articles/natur...
00:00 Description
00:12 1 hour of training
00:59 2 hours of training
01:35 4 hours of training
02:37 8 hours of training
03:44 16 hours of training
05:02 32 hours of training
Experimental setup:
OS: Fedora Linux 42 (Workstation Edition) x86_64
CPU: AMD Ryzen 5 2600X (12) @ 3.60 GHz
GPU: NVIDIA GeForce RTX 3060 ti (8GB VRAM)
RAM: 32 GB DDR4 3200 MHz
Training took ~32 hours on an RTX 3060ti (8GB VRAM). I trained DQN for 50 million environment steps using a replay buffer of 500,000 transitions (RAM limited :D), while the original DeepMind setup used 1,000,000.
Citation:
Mnih, V., Kavukcuoglu, K., Silver, D. et al. Human-level control through deep reinforcement learning. Nature 518, 529–533 (2015). https://doi.org/10.1038/nature14236
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
