{"id":13794265,"url":"https://github.com/miyosuda/async_deep_reinforce","last_synced_at":"2026-04-06T02:20:43.946Z","repository":{"id":54216863,"uuid":"55909050","full_name":"miyosuda/async_deep_reinforce","owner":"miyosuda","description":"Asynchronous Methods for Deep Reinforcement Learning","archived":false,"fork":false,"pushed_at":"2018-08-09T09:29:30.000Z","size":306,"stargazers_count":590,"open_issues_count":36,"forks_count":194,"subscribers_count":49,"default_branch":"master","last_synced_at":"2024-08-03T23:03:32.554Z","etag":null,"topics":["a3c","deep-learning","reinforcement-learning","tensorflow"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/miyosuda.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE.txt","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null}},"created_at":"2016-04-10T16:10:40.000Z","updated_at":"2024-07-28T08:06:16.000Z","dependencies_parsed_at":"2022-08-13T09:21:04.333Z","dependency_job_id":null,"html_url":"https://github.com/miyosuda/async_deep_reinforce","commit_stats":null,"previous_names":[],"tags_count":1,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/miyosuda%2Fasync_deep_reinforce","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/miyosuda%2Fasync_deep_reinforce/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/miyosuda%2Fasync_deep_reinforce/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/miyosuda%2Fasync_deep_reinforce/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/miyosuda","download_url":"https://codeload.github.com/miyosuda/async_deep_reinforce/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":225152702,"owners_count":17429179,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["a3c","deep-learning","reinforcement-learning","tensorflow"],"created_at":"2024-08-03T23:00:37.899Z","updated_at":"2026-04-06T02:20:43.919Z","avatar_url":"https://github.com/miyosuda.png","language":"Python","funding_links":[],"categories":["Machine Learning"],"sub_categories":["Reinforcement Learning"],"readme":"# async_deep_reinforce\n\nAsynchronous deep reinforcement learning\n\n## About\n\nAn attempt to repdroduce Google Deep Mind's paper \"Asynchronous Methods for Deep Reinforcement Learning.\"\n\nhttp://arxiv.org/abs/1602.01783\n\nAsynchronous Advantage Actor-Critic (A3C) method for playing \"Atari Pong\" is implemented with TensorFlow.\nBoth A3C-FF and A3C-LSTM are implemented.\n\nLearning result movment after 26 hours (A3C-FF) is like this.\n\n[![Learning result after 26 hour](http://narr.jp/private/miyoshi/deep_learning/a3c_preview_image.jpg)](https://youtu.be/ZU71YdAedZs)\n\nAny advice or suggestion is strongly welcomed in issues thread.\n\nhttps://github.com/miyosuda/async_deep_reinforce/issues/1\n\n## How to build\n\nFirst we need to build multi thread ready version of Arcade Learning Enviroment.\nI made some modification to it to run it on multi thread enviroment.\n\n    $ git clone https://github.com/miyosuda/Arcade-Learning-Environment.git\n    $ cd Arcade-Learning-Environment\n    $ cmake -DUSE_SDL=ON -DUSE_RLGLUE=OFF -DBUILD_EXAMPLES=OFF .\n    $ make -j 4\n\t\n    $ pip install .\n\nI recommend to install it on VirtualEnv environment.\n\n## How to run\n\nTo train,\n\n    $python a3c.py\n\nTo display the result with game play,\n\n    $python a3c_disp.py\n\n## Using GPU\nTo enable gpu, change \"USE_GPU\" flag in \"constants.py\".\n\nWhen running with 8 parallel game environemts, speeds of GPU (GTX980Ti) and CPU(Core i7 6700) were like this. (Recorded with LOCAL_T_MAX=20 setting.)\n\n|type | A3C-FF             |A3C-LSTM          |\n|-----|--------------------|------------------|\n| GPU | 1722 steps per sec |864 steps per sec |\n| CPU | 1077 steps per sec |540 steps per sec |\n\n\n## Result\nScore plots of local threads of pong were like these. (with GTX980Ti)\n\n### A3C-LSTM LOCAL_T_MAX = 5\n\n![A3C-LSTM T=5](./docs/graph_t5.png)\n\n### A3C-LSTM LOCAL_T_MAX = 20\n\n![A3C-LSTM T=20](./docs/graph_t20.png)\n\nScores are not averaged using global network unlike the original paper.\n\n## Requirements\n- TensorFlow r1.0\n- numpy\n- cv2\n- matplotlib\n\n## References\n\nThis project uses setting written in muupan's wiki [muuupan/async-rl] (https://github.com/muupan/async-rl/wiki)\n\n\n## Acknowledgements\n\n- [@aravindsrinivas](https://github.com/aravindsrinivas) for providing information for some of the hyper parameters.\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmiyosuda%2Fasync_deep_reinforce","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fmiyosuda%2Fasync_deep_reinforce","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmiyosuda%2Fasync_deep_reinforce/lists"}