{"id":13454799,"url":"https://github.com/devsisters/DQN-tensorflow","last_synced_at":"2025-03-24T07:32:11.606Z","repository":{"id":40655558,"uuid":"58858323","full_name":"devsisters/DQN-tensorflow","owner":"devsisters","description":"Tensorflow implementation of Human-Level Control through Deep Reinforcement Learning","archived":false,"fork":false,"pushed_at":"2019-04-18T18:36:45.000Z","size":30282,"stargazers_count":2522,"open_issues_count":38,"forks_count":765,"subscribers_count":141,"default_branch":"master","last_synced_at":"2025-03-24T03:01:34.465Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/devsisters.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null}},"created_at":"2016-05-15T11:33:47.000Z","updated_at":"2025-03-23T02:55:16.000Z","dependencies_parsed_at":"2022-07-14T04:50:33.729Z","dependency_job_id":null,"html_url":"https://github.com/devsisters/DQN-tensorflow","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/devsisters%2FDQN-tensorflow","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/devsisters%2FDQN-tensorflow/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/devsisters%2FDQN-tensorflow/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/devsisters%2FDQN-tensorflow/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/devsisters","download_url":"https://codeload.github.com/devsisters/DQN-tensorflow/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":245227539,"owners_count":20580895,"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":[],"created_at":"2024-07-31T08:00:58.040Z","updated_at":"2025-03-24T07:32:11.579Z","avatar_url":"https://github.com/devsisters.png","language":"Python","funding_links":[],"categories":["Models/Projects","Python (144)","Python","模型项目","Machine Learning","Meta-Learning"],"sub_categories":["微信群","Reinforcement Learning","Multi-task learning"],"readme":"# Human-Level Control through Deep Reinforcement Learning\n\nTensorflow implementation of [Human-Level Control through Deep Reinforcement Learning](http://home.uchicago.edu/~arij/journalclub/papers/2015_Mnih_et_al.pdf).\n\n![model](assets/model.png)\n\nThis implementation contains:\n\n1. Deep Q-network and Q-learning\n2. Experience replay memory\n    - to reduce the correlations between consecutive updates\n3. Network for Q-learning targets are fixed for intervals\n    - to reduce the correlations between target and predicted Q-values\n\n\n## Requirements\n\n- Python 2.7 or Python 3.3+\n- [gym](https://github.com/openai/gym)\n- [tqdm](https://github.com/tqdm/tqdm)\n- [SciPy](http://www.scipy.org/install.html) or [OpenCV2](http://opencv.org/)\n- [TensorFlow 0.12.0](https://github.com/tensorflow/tensorflow/tree/r0.12)\n\n\n## Usage\n\nFirst, install prerequisites with:\n\n    $ pip install tqdm gym[all]\n\nTo train a model for Breakout:\n\n    $ python main.py --env_name=Breakout-v0 --is_train=True\n    $ python main.py --env_name=Breakout-v0 --is_train=True --display=True\n\nTo test and record the screen with gym:\n\n    $ python main.py --is_train=False\n    $ python main.py --is_train=False --display=True\n\n\n## Results\n\nResult of training for 24 hours using GTX 980 ti.\n\n![best](assets/best.gif)\n\n\n## Simple Results\n\nDetails of `Breakout` with model `m2`(red) for 30 hours using GTX 980 Ti.\n\n![tensorboard](assets/0620_scalar_step_m2.png)\n\nDetails of `Breakout` with model `m3`(red) for 30 hours using GTX 980 Ti.\n\n![tensorboard](assets/0620_scalar_step_m3.png)\n\n\n## Detailed Results\n\n**[1] Action-repeat (frame-skip) of 1, 2, and 4 without learning rate decay**\n\n![A1_A2_A4_0.00025lr](assets/A1_A2_A4_0.00025lr.png)\n\n**[2] Action-repeat (frame-skip) of 1, 2, and 4 with learning rate decay**\n\n![A1_A2_A4_0.0025lr](assets/A1_A2_A4_0.0025lr.png)\n\n**[1] \u0026 [2]**\n\n![A1_A2_A4_0.00025lr_0.0025lr](assets/A1_A2_A4_0.00025lr_0.0025lr.png)\n\n\n**[3] Action-repeat of 4 for DQN (dark blue) Dueling DQN (dark green) DDQN (brown) Dueling DDQN (turquoise)**\n\nThe current hyper parameters and gradient clipping are not implemented as it is in the paper.\n\n![A4_duel_double](assets/A4_duel_double.png)\n\n\n**[4] Distributed action-repeat (frame-skip) of 1 without learning rate decay**\n\n![A1_0.00025lr_distributed](assets/A4_0.00025lr_distributed.png)\n\n**[5] Distributed action-repeat (frame-skip) of 4 without learning rate decay**\n\n![A4_0.00025lr_distributed](assets/A4_0.00025lr_distributed.png)\n\n\n## References\n\n- [simple_dqn](https://github.com/tambetm/simple_dqn.git)\n- [Code for Human-level control through deep reinforcement learning](https://sites.google.com/a/deepmind.com/dqn/)\n\n\n## License\n\nMIT License.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdevsisters%2FDQN-tensorflow","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fdevsisters%2FDQN-tensorflow","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdevsisters%2FDQN-tensorflow/lists"}