[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-microsoft--maro":3,"tool-microsoft--maro":61},[4,18,26,36,44,53],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":10,"last_commit_at":24,"category_tags":25,"status":17},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,"2026-04-05T11:01:52",[14,15,13],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"status":17},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",150037,2,"2026-04-10T23:33:47",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"status":17},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",108322,"2026-04-10T11:39:34",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":32,"last_commit_at":59,"category_tags":60,"status":17},4721,"markitdown","microsoft\u002Fmarkitdown","MarkItDown 是一款由微软 AutoGen 团队打造的轻量级 Python 工具，专为将各类文件高效转换为 Markdown 格式而设计。它支持 PDF、Word、Excel、PPT、图片（含 OCR）、音频（含语音转录）、HTML 乃至 YouTube 链接等多种格式的解析，能够精准提取文档中的标题、列表、表格和链接等关键结构信息。\n\n在人工智能应用日益普及的今天，大语言模型（LLM）虽擅长处理文本，却难以直接读取复杂的二进制办公文档。MarkItDown 恰好解决了这一痛点，它将非结构化或半结构化的文件转化为模型“原生理解”且 Token 效率极高的 Markdown 格式，成为连接本地文件与 AI 分析 pipeline 的理想桥梁。此外，它还提供了 MCP（模型上下文协议）服务器，可无缝集成到 Claude Desktop 等 LLM 应用中。\n\n这款工具特别适合开发者、数据科学家及 AI 研究人员使用，尤其是那些需要构建文档检索增强生成（RAG）系统、进行批量文本分析或希望让 AI 助手直接“阅读”本地文件的用户。虽然生成的内容也具备一定可读性，但其核心优势在于为机器",93400,"2026-04-06T19:52:38",[52,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":66,"readme_en":67,"readme_zh":68,"quickstart_zh":69,"use_case_zh":70,"hero_image_url":71,"owner_login":72,"owner_name":73,"owner_avatar_url":74,"owner_bio":75,"owner_company":76,"owner_location":76,"owner_email":77,"owner_twitter":78,"owner_website":79,"owner_url":80,"languages":81,"stars":121,"forks":122,"last_commit_at":123,"license":124,"difficulty_score":125,"env_os":126,"env_gpu":127,"env_ram":127,"env_deps":128,"category_tags":134,"github_topics":136,"view_count":32,"oss_zip_url":76,"oss_zip_packed_at":76,"status":17,"created_at":152,"updated_at":153,"faqs":154,"releases":184},6415,"microsoft\u002Fmaro","maro","Multi-Agent Resource Optimization (MARO) platform is an instance of Reinforcement Learning as a Service (RaaS) for real-world resource optimization problems. ","MARO（Multi-Agent Resource Optimization）是一个专为现实世界资源优化问题打造的强化学习服务平台。它旨在解决物流、云计算等工业场景中复杂的资源调度难题，例如集装箱库存管理、共享单车重新分配以及虚拟机动态调度等。在这些场景中，传统算法往往难以应对大规模、动态变化的环境，而 MARO 通过多智能体强化学习技术，让多个决策主体协同工作，从而找到更高效的资源分配策略。\n\n这款平台非常适合人工智能研究人员、算法工程师以及需要优化资源调度系统的开发者使用。无论是希望验证新算法的学术界人士，还是致力于提升运营效率的企业技术团队，都能从中受益。MARO 的独特亮点在于其“强化学习即服务”（RaaS）的设计理念，提供了标准化的场景接口和完整的训练评估流程，让用户无需从零构建仿真环境即可快速上手。此外，它还支持 Docker 部署和多种主流 Python 版本，具备良好的可扩展性与易用性，帮助用户将前沿的 AI 理论迅速转化为解决实际问题的生产力。","[![License](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fl\u002Fpymaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fblob\u002Fmaster\u002FLICENSE)\n[![Platform](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fplatform.svg)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F)\n[![Python Versions](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fpyversions\u002Fpymaro.svg?logo=python&logoColor=white)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)\n[![Code Size](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flanguages\u002Fcode-size\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro)\n[![Docker Size](https:\u002F\u002Fimg.shields.io\u002Fdocker\u002Fimage-size\u002Fmaro2020\u002Fmaro)](https:\u002F\u002Fhub.docker.com\u002Frepository\u002Fdocker\u002Fmaro2020\u002Fmaro\u002Ftags?page=1)\n[![Issues](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fissues\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fissues)\n[![Pull Requests](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fissues-pr\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpulls)\n[![Dependencies](https:\u002F\u002Fimg.shields.io\u002Flibrariesio\u002Fgithub\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Flibraries.io\u002Fpypi\u002Fpymaro)\n[![test](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Ftest\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Factions?query=workflow%3Atest)\n[![build](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Fbuild\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Factions?query=workflow%3Abuild)\n[![docker](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Fdocker\u002Fbadge.svg)](https:\u002F\u002Fhub.docker.com\u002Frepository\u002Fdocker\u002Fmaro2020\u002Fmaro)\n[![docs](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_6bf48b3e9a6d.png)](https:\u002F\u002Fmaro.readthedocs.io\u002F)\n[![PypI Versions](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fpymaro)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)\n[![Wheel](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fwheel\u002Fpymaro)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)\n[![Citi Bike](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fciti_bike.svg)](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fciti_bike.html)\n[![CIM](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fcim.svg)](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fcontainer_inventory_management.html)\n[![VM Scheduling](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fvm_scheduling.svg)](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fvm_scheduling.html)\n[![Gitter](https:\u002F\u002Fimg.shields.io\u002Fgitter\u002Froom\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgitter.im\u002FMicrosoft\u002FMARO#)\n[![Stack Overflow](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fstack_overflow.svg)](https:\u002F\u002Fstackoverflow.com\u002Fquestions\u002Fask?tags=maro)\n[![Releases](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Frelease-date-pre\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Freleases)\n[![Commits](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fcommits-since\u002Fmicrosoft\u002Fmaro\u002Flatest\u002Fmaster)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fcommits\u002Fmaster)\n[![Vulnerability Scan](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Fvulnerability%20scan\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Factions?query=workflow%3A%22vulnerability+scan%22)\n[![Lint](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Flint\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Factions?query=workflow%3Alint)\n[![Coverage](https:\u002F\u002Fimg.shields.io\u002Fcodecov\u002Fc\u002Fgithub\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fcodecov.io\u002Fgh\u002Fmicrosoft\u002Fmaro)\n[![Downloads](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fdm\u002Fpymaro)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)\n[![Docker Pulls](https:\u002F\u002Fimg.shields.io\u002Fdocker\u002Fpulls\u002Fmaro2020\u002Fmaro)](https:\u002F\u002Fhub.docker.com\u002Frepository\u002Fdocker\u002Fmaro2020\u002Fmaro)\n[![Play with MARO](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fplay_with_maro.svg)](https:\u002F\u002Fhub.docker.com\u002Fr\u002Fmaro2020\u002Fmaro)\n\n# [![MARO LOGO](.\u002Fdocs\u002Fsource\u002Fimages\u002Flogo.svg)](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002F)\n\nMulti-Agent Resource Optimization (MARO) platform is an instance of Reinforcement\nlearning as a Service (RaaS) for real-world resource optimization. It can be\napplied to many important industrial domains, such as [container inventory\nmanagement](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fcontainer_inventory_management.html)\nin logistics, [bike repositioning](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fciti_bike.html)\nin transportation, [virtual machine](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fvm_scheduling.html) provisioning in data centers, and asset management in finance. Besides\n[Reinforcement Learning](https:\u002F\u002Fwww.andrew.cmu.edu\u002Fcourse\u002F10-703\u002Ftextbook\u002FBartoSutton.pdf) (RL),\nit also supports other planning\u002Fdecision mechanisms, such as\n[Operations Research](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FOperations_research).\n\nKey Components of MARO:\n\n- Simulation toolkit: it provides some predefined scenarios, and the reusable\nwheels for building new scenarios.\n- RL toolkit: it provides a full-stack abstraction for RL, such as agent manager,\nagent, RL algorithms, learner, actor, and various shapers.\n- Distributed toolkit: it provides distributed communication components, interface\nof user-defined functions for message auto-handling, cluster provision, and job orchestration.\n\n![MARO Key Components](.\u002Fdocs\u002Fsource\u002Fimages\u002Fmaro_overview.svg)\n\n## Contents\n\n| File\u002Ffolder | Description                                                                                       |\n| ----------- | ------------------------------------------------------------------------------------------------- |\n| `maro`      | MARO source code.                                                                                 |\n| `docs`      | MARO docs, it is host on [readthedocs](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002F).                   |\n| `examples`  | Showcase of MARO.                                                                                 |\n| `notebooks` | MARO quick-start notebooks.                                                                       |\n\n*Try [MARO playground](#run-playground) to have a quick experience.*\n\n## Install MARO from [PyPI](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)\n\n*Notes: The CLI commands (including the visualization tool) are not included in pymaro package. To enable these support, you need to install from source.*\n\n- Mac OS \u002F Linux\n\n  ```sh\n  pip install pymaro\n  ```\n\n- Windows\n\n  ```powershell\n  # Install torch first, if you don't have one.\n  pip install torch===1.6.0 torchvision===0.7.0 -f https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Ftorch_stable.html\n\n  pip install pymaro\n  ```\n\n## Install MARO from Source\n\n*Notes: Install from source if you want to use the CLI commands (including the visualization tool).*\n\n- Prerequisites\n  - C++ Compiler\n    - Linux or Mac OS X: `gcc`\n    - Windows: [Build Tools for Visual Studio 2017](https:\u002F\u002Fvisualstudio.microsoft.com\u002Fthank-you-downloading-visual-studio\u002F?sku=BuildTools&rel=15)\n\n- Enable Virtual Environment\n  - Mac OS \u002F Linux\n\n    ```sh\n    # If your environment is not clean, create a virtual environment firstly.\n    python -m venv maro_venv\n    source .\u002Fmaro_venv\u002Fbin\u002Factivate\n    ```\n\n  - Windows\n\n    ```powershell\n    # If your environment is not clean, create a virtual environment firstly.\n    python -m venv maro_venv\n\n    # You may need this for SecurityError in PowerShell.\n    Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy Unrestricted\n\n    # Activate the virtual environment.\n    .\\maro_venv\\Scripts\\activate\n    ```\n\n- Install MARO\n\n  ```sh\n  # Git Clone the whole source code.\n  git clone https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro.git\n  ```\n\n  - Mac OS \u002F Linux\n\n    ```sh\n    # Install MARO from source.\n    bash scripts\u002Finstall_maro.sh;\n    pip install -r .\u002Frequirements.dev.txt;\n    ```\n\n  - Windows\n\n    ```powershell\n    # Install MARO from source.\n    .\\scripts\\install_maro.bat;\n    pip install -r .\u002Frequirements.dev.txt;\n    ```\n\n- *Notes: If your package is not found, remember to set your PYTHONPATH*\n\n  - Mac OS \u002F Linux\n\n  ```sh\n  export PYTHONPATH=PATH-TO-MARO\n  ```\n\n  - Windows\n\n  ```powershell\n  $Env:PYTHONPATH=PATH-TO-MARO\n  ```\n\n## Quick Example\n\n```python\nfrom maro.simulator import Env\n\nenv = Env(scenario=\"cim\", topology=\"toy.5p_ssddd_l0.0\", start_tick=0, durations=100)\n\nmetrics, decision_event, is_done = env.step(None)\n\nwhile not is_done:\n    metrics, decision_event, is_done = env.step(None)\n\nprint(f\"environment metrics: {env.metrics}\")\n\n```\n\n## [Environment Visualization](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002F)\n\n```sh\n# Enable environment dump feature, when initializing the environment instance\nenv = Env(scenario=\"cim\",\n          topology=\"toy.5p_ssddd_l0.0\",\n          start_tick=0,\n          durations=100,\n          options={\"enable-dump-snapshot\": \".\u002Fdump_data\"})\n\n# Inspect environment with the dump data\nmaro inspector dashboard --source_path .\u002Fdump_data\u002FYOUR_SNAPSHOT_DUMP_FOLDER\n```\n\n### Show Cases\n\n- Case I - Container Inventory Management\n![CIM Inter Epoch](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_61d56d65e1ec.gif)\n![CIM Intra Epoch](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_9b99ba5a8266.gif)\n\n- Case II - Citi Bike\n![Citi Bike Inter Epoch](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_e228088a4901.gif)\n![Citi Bike Intra Epoch](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_9b951f4b263f.gif)\n\n## Run Playground\n\n- Pull from [Docker Hub](https:\u002F\u002Fhub.docker.com\u002Fr\u002Fmaro2020\u002Fplayground)\n\n  ```sh\n  # Pull the docker image from docker hub\n  docker pull maro2020\u002Fplayground\n\n  # Run playground container.\n  # Redis commander (GUI for redis) -> http:\u002F\u002F127.0.0.1:40009\n  # Jupyter lab with maro -> http:\u002F\u002F127.0.0.1:40010\n  docker run -p 40009:40009 -p 40010:40010 maro2020\u002Fplayground\n  ```\n\n- Build from source\n  - Mac OS \u002F Linux\n\n    ```sh\n    # Build playground image.\n    bash .\u002Fscripts\u002Fbuild_playground.sh\n\n    # Run playground container.\n    # Redis commander (GUI for redis) -> http:\u002F\u002F127.0.0.1:40009\n    # Jupyter lab with maro -> http:\u002F\u002F127.0.0.1:40010\n    docker run -p 40009:40009 -p 40010:40010 maro2020\u002Fplayground\n    ```\n\n  - Windows\n\n    ```powershell\n    # Build playground image.\n    .\\scripts\\build_playground.bat\n\n    # Run playground container.\n    # Redis commander (GUI for redis) -> http:\u002F\u002F127.0.0.1:40009\n    # Jupyter lab with maro -> http:\u002F\u002F127.0.0.1:40010\n    docker run -p 40009:40009 -p 40010:40010 maro2020\u002Fplayground\n    ```\n\n## Contributing\n\nThis project welcomes contributions and suggestions. Most contributions require\nyou to agree to a Contributor License Agreement (CLA) declaring that you have\nthe right to, and actually do, grant us the rights to use your contribution. For\ndetails, visit https:\u002F\u002Fcla.opensource.microsoft.com.\n\nWhen you submit a pull request, a CLA bot will automatically determine whether\nyou need to provide a CLA and decorate the PR appropriately (e.g., status check,\ncomment). Simply follow the instructions provided by the bot. You will only need\nto do this once across all repos using our CLA.\n\nThis project has adopted the\n[Microsoft Open Source Code of Conduct](https:\u002F\u002Fopensource.microsoft.com\u002Fcodeofconduct\u002F).\nFor more information see the\n[Code of Conduct FAQ](https:\u002F\u002Fopensource.microsoft.com\u002Fcodeofconduct\u002Ffaq\u002F)\nor contact [opencode@microsoft.com](mailto:opencode@microsoft.com)\nwith any additional questions or comments.\n\n## Related Papers\n\n### [Container Inventory Management](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fcontainer_inventory_management.html)\n\n![CIM Vis](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_93623f5b78c3.gif)\n\nWenlei Shi, Xinran Wei, Jia Zhang, Xiaoyuan Ni, Arthur Jiang, Jiang Bian, Tie-Yan Liu. \"[Cooperative Policy Learning with Pre-trained Heterogeneous Observation Representations](https:\u002F\u002Farxiv.org\u002Fabs\u002F2012.13099)\". AAMAS 2021\n\nXihan Li, Jia Zhang, Jiang Bian, Yunhai Tong, Tie-Yan Liu. \"[A Cooperative Multi-Agent Reinforcement Learning Framework for Resource Balancing in Complex Logistics Network](https:\u002F\u002Farxiv.org\u002Fabs\u002F1903.00714)\". AAMAS 2019\n\n## Related News\n\n[MSRA Top-10 Hack-Techs in 2021](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FY4kjQq8gKcsEsVadjdwnEQ)\n\n[Open Source Platform MARO: Anywhere Door for Resource Optimization](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FzXIpgzomLhDWS_YUFmRlEQ)\n\n[AI from \"Point\" to \"Surface\"](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FPggO49wwmS7grTO0nEMgVQ)\n\n## [Cite Us](.\u002FCITATION)\n\n## License\n\nCopyright (c) Microsoft Corporation. All rights reserved.\n\nLicensed under the [MIT](.\u002FLICENSE) License.\n","[![许可证](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fl\u002Fpymaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fblob\u002Fmaster\u002FLICENSE)\n[![平台](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fplatform.svg)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F)\n[![Python版本](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fpyversions\u002Fpymaro.svg?logo=python&logoColor=white)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)\n[![代码量](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flanguages\u002Fcode-size\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro)\n[![Docker镜像大小](https:\u002F\u002Fimg.shields.io\u002Fdocker\u002Fimage-size\u002Fmaro2020\u002Fmaro)](https:\u002F\u002Fhub.docker.com\u002Frepository\u002Fdocker\u002Fmaro2020\u002Fmaro\u002Ftags?page=1)\n[![问题](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fissues\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fissues)\n[![拉取请求](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fissues-pr\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpulls)\n[![依赖](https:\u002F\u002Fimg.shields.io\u002Flibrariesio\u002Fgithub\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Flibraries.io\u002Fpypi\u002Fpymaro)\n[![测试](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Ftest\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Factions?query=workflow%3Atest)\n[![构建](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Fbuild\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Factions?query=workflow%3Abuild)\n[![Docker](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Fdocker\u002Fbadge.svg)](https:\u002F\u002Fhub.docker.com\u002Frepository\u002Fdocker\u002Fmaro2020\u002Fmaro)\n[![文档](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_6bf48b3e9a6d.png)](https:\u002F\u002Fmaro.readthedocs.io\u002F)\n[![PyPI版本](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fpymaro)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)\n[![轮子包](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fwheel\u002Fpymaro)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)\n[![Citi Bike](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fciti_bike.svg)](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fciti_bike.html)\n[![CIM](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fcim.svg)](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fcontainer_inventory_management.html)\n[![虚拟机调度](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fvm_scheduling.svg)](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fvm_scheduling.html)\n[![Gitter](https:\u002F\u002Fimg.shields.io\u002Fgitter\u002Froom\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgitter.im\u002FMicrosoft\u002FMARO#)\n[![Stack Overflow](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fstack_overflow.svg)](https:\u002F\u002Fstackoverflow.com\u002Fquestions\u002Fask?tags=maro)\n[![发布](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Frelease-date-pre\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Freleases)\n[![提交](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fcommits-since\u002Fmicrosoft\u002Fmaro\u002Flatest\u002Fmaster)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fcommits\u002Fmaster)\n[![漏洞扫描](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Fvulnerability%20scan\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Factions?query=workflow%3A%22vulnerability+scan%22)\n[![代码风格检查](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fworkflows\u002Flint\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Factions?query=workflow%3Alint)\n[![覆盖率](https:\u002F\u002Fimg.shields.io\u002Fcodecov\u002Fc\u002Fgithub\u002Fmicrosoft\u002Fmaro)](https:\u002F\u002Fcodecov.io\u002Fgh\u002Fmicrosoft\u002Fmaro)\n[![下载量](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fdm\u002Fpymaro)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)\n[![Docker拉取次数](https:\u002F\u002Fimg.shields.io\u002Fdocker\u002Fpulls\u002Fmaro2020\u002Fmaro)](https:\u002F\u002Fhub.docker.com\u002Frepository\u002Fdocker\u002Fmaro2020\u002Fmaro)\n[![与MARO一起玩](https:\u002F\u002Fraw.githubusercontent.com\u002Fmicrosoft\u002Fmaro\u002Fmaster\u002Fdocs\u002Fsource\u002Fimages\u002Fbadges\u002Fplay_with_maro.svg)](https:\u002F\u002Fhub.docker.com\u002Fr\u002Fmaro2020\u002Fmaro)\n\n# [![MARO LOGO](.\u002Fdocs\u002Fsource\u002Fimages\u002Flogo.svg)](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002F)\n\n多智能体资源优化（MARO）平台是面向现实世界资源优化的强化学习即服务（RaaS）实例。它可以应用于许多重要的工业领域，例如物流中的[集装箱库存管理](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fcontainer_inventory_management.html)、交通领域的[共享单车调度](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fciti_bike.html)、数据中心的[虚拟机调度](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fvm_scheduling.html)以及金融领域的资产管理等。除了[强化学习](https:\u002F\u002Fwww.andrew.cmu.edu\u002Fcourse\u002F10-703\u002Ftextbook\u002FBartoSutton.pdf)之外，它还支持其他规划\u002F决策机制，如[运筹学](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FOperations_research)。\n\nMARO的主要组件：\n\n- 模拟工具包：提供一些预定义场景，并包含用于构建新场景的可重用模块。\n- 强化学习工具包：为强化学习提供全栈抽象，包括智能体管理器、智能体、强化学习算法、学习者、执行者以及各种策略调整器。\n- 分布式工具包：提供分布式通信组件、用户自定义函数接口以实现消息自动处理、集群部署和作业编排等功能。\n\n![MARO主要组件](.\u002Fdocs\u002Fsource\u002Fimages\u002Fmaro_overview.svg)\n\n## 目录\n\n| 文件\u002F文件夹 | 描述                                                                                       |\n| ----------- | ------------------------------------------------------------------------------------------------- |\n| `maro`      | MARO源代码。                                                                                 |\n| `docs`      | MARO文档，托管在[readthedocs](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002F)上。                   |\n| `examples`  | MARO示例展示。                                                                                 |\n| `notebooks` | MARO快速入门笔记本。                                                                       |\n\n*尝试[MARO游乐场](#run-playground)，快速体验一下吧。*\n\n## 从[PyPI](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpymaro\u002F#files)安装MARO\n\n*注：CLI命令（包括可视化工具）并未包含在pymaro软件包中。若需启用这些功能，需从源码安装。*\n\n- Mac OS \u002F Linux\n\n  ```sh\n  pip install pymaro\n  ```\n\n- Windows\n\n  ```powershell\n  # 如果尚未安装，请先安装torch。\n  pip install torch===1.6.0 torchvision===0.7.0 -f https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Ftorch_stable.html\n\n  pip install pymaro\n  ```\n\n## 从源代码安装 MARO\n\n*注：如果您想使用 CLI 命令（包括可视化工具），请从源代码安装。*\n\n- 先决条件\n  - C++ 编译器\n    - Linux 或 Mac OS X：`gcc`\n    - Windows：[适用于 Visual Studio 2017 的构建工具](https:\u002F\u002Fvisualstudio.microsoft.com\u002Fthank-you-downloading-visual-studio\u002F?sku=BuildTools&rel=15)\n\n- 启用虚拟环境\n  - Mac OS \u002F Linux\n\n    ```sh\n    # 如果您的环境不干净，请先创建一个虚拟环境。\n    python -m venv maro_venv\n    source .\u002Fmaro_venv\u002Fbin\u002Factivate\n    ```\n\n  - Windows\n\n    ```powershell\n    # 如果您的环境不干净，请先创建一个虚拟环境。\n    python -m venv maro_venv\n\n    # 在 PowerShell 中可能会遇到 SecurityError，您可能需要执行以下命令。\n    Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy Unrestricted\n\n    # 激活虚拟环境。\n    .\\maro_venv\\Scripts\\activate\n    ```\n\n- 安装 MARO\n\n  ```sh\n  # 使用 Git 克隆整个源代码。\n  git clone https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro.git\n  ```\n\n  - Mac OS \u002F Linux\n\n    ```sh\n    # 从源代码安装 MARO。\n    bash scripts\u002Finstall_maro.sh;\n    pip install -r .\u002Frequirements.dev.txt;\n    ```\n\n  - Windows\n\n    ```powershell\n    # 从源代码安装 MARO。\n    .\\scripts\\install_maro.bat;\n    pip install -r .\u002Frequirements.dev.txt;\n    ```\n\n- *注：如果找不到您的包，请记得设置 PYTHONPATH*\n\n  - Mac OS \u002F Linux\n\n  ```sh\n  export PYTHONPATH=MARO路径\n  ```\n\n  - Windows\n\n  ```powershell\n  $Env:PYTHONPATH=MARO路径\n  ```\n\n## 快速示例\n\n```python\nfrom maro.simulator import Env\n\nenv = Env(scenario=\"cim\", topology=\"toy.5p_ssddd_l0.0\", start_tick=0, durations=100)\n\nmetrics, decision_event, is_done = env.step(None)\n\nwhile not is_done:\n    metrics, decision_event, is_done = env.step(None)\n\nprint(f\"environment metrics: {env.metrics}\")\n\n```\n\n## [环境可视化](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002F)\n\n```sh\n# 在初始化环境实例时启用环境转储功能\nenv = Env(scenario=\"cim\",\n          topology=\"toy.5p_ssddd_l0.0\",\n          start_tick=0,\n          durations=100,\n          options={\"enable-dump-snapshot\": \".\u002Fdump_data\"})\n\n# 使用转储数据检查环境\nmaro inspector dashboard --source_path .\u002Fdump_data\u002F您的快照转储文件夹\n```\n\n### 案例展示\n\n- 案例 I - 集装箱库存管理\n![CIM 跨周期](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_61d56d65e1ec.gif)\n![CIM 周期内](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_9b99ba5a8266.gif)\n\n- 案例 II - Citi Bike\n![Citi Bike 跨周期](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_e228088a4901.gif)\n![Citi Bike 周期内](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_9b951f4b263f.gif)\n\n## 运行 Playground\n\n- 从 [Docker Hub](https:\u002F\u002Fhub.docker.com\u002Fr\u002Fmaro2020\u002Fplayground) 拉取镜像\n\n  ```sh\n  # 从 Docker Hub 拉取镜像\n  docker pull maro2020\u002Fplayground\n\n  # 运行 Playground 容器。\n  # Redis commander（Redis 的 GUI）-> http:\u002F\u002F127.0.0.1:40009\n  # 带有 MARO 的 Jupyter Lab -> http:\u002F\u002F127.0.0.1:40010\n  docker run -p 40009:40009 -p 40010:40010 maro2020\u002Fplayground\n  ```\n\n- 从源代码构建\n  - Mac OS \u002F Linux\n\n    ```sh\n    # 构建 Playground 镜像。\n    bash .\u002Fscripts\u002Fbuild_playground.sh\n\n    # 运行 Playground 容器。\n    # Redis commander（Redis 的 GUI）-> http:\u002F\u002F127.0.0.1:40009\n    # 带有 MARO 的 Jupyter Lab -> http:\u002F\u002F127.0.0.1:40010\n    docker run -p 40009:40009 -p 40010:40010 maro2020\u002Fplayground\n    ```\n\n  - Windows\n\n    ```powershell\n    # 构建 Playground 镜像。\n    .\\scripts\\build_playground.bat\n\n    # 运行 Playground 容器。\n    # Redis commander（Redis 的 GUI）-> http:\u002F\u002F127.0.0.1:40009\n    # 带有 MARO 的 Jupyter Lab -> http:\u002F\u002F127.0.0.1:40010\n    docker run -p 40009:40009 -p 40010:40010 maro2020\u002Fplayground\n    ```\n\n## 贡献\n本项目欢迎贡献和建议。大多数贡献都需要您同意贡献者许可协议 (CLA)，声明您有权并确实授予我们使用您贡献的权利。有关详细信息，请访问 https:\u002F\u002Fcla.opensource.microsoft.com。\n\n当您提交拉取请求时，CLA 机器人会自动确定您是否需要提供 CLA，并相应地标记 PR（例如状态检查、评论）。只需按照机器人提供的指示操作即可。对于使用我们 CLA 的所有仓库，您只需执行一次此操作。\n\n该项目已采用\n[微软开源行为准则](https:\u002F\u002Fopensource.microsoft.com\u002Fcodeofconduct\u002F)。\n更多信息请参阅\n[行为准则常见问题解答](https:\u002F\u002Fopensource.microsoft.com\u002Fcodeofconduct\u002Ffaq\u002F)\n或发送电子邮件至 [opencode@microsoft.com](mailto:opencode@microsoft.com)\n以获取更多问题或意见。\n\n## 相关论文\n\n### [集装箱库存管理](https:\u002F\u002Fmaro.readthedocs.io\u002Fen\u002Flatest\u002Fscenarios\u002Fcontainer_inventory_management.html)\n\n![CIM 可视化](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_readme_93623f5b78c3.gif)\n\nWenlei Shi, Xinran Wei, Jia Zhang, Xiaoyuan Ni, Arthur Jiang, Jiang Bian, Tie-Yan Liu. “[具有预训练异构观测表示的协作策略学习](https:\u002F\u002Farxiv.org\u002Fabs\u002F2012.13099)”。AAMAS 2021\n\nXihan Li, Jia Zhang, Jiang Bian, Yunhai Tong, Tie-Yan Liu. “[用于复杂物流网络中资源平衡的协作式多智能体强化学习框架](https:\u002F\u002Farxiv.org\u002Fabs\u002F1903.00714)”。AAMAS 2019\n\n## 相关新闻\n\n[MSRA 2021 年十大黑客技术](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FY4kjQq8gKcsEsVadjdwnEQ)\n\n[开源平台 MARO：资源优化的任意门](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FzXIpgzomLhDWS_YUFmRlEQ)\n\n[AI 从“点”到“面”](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FPggO49wwmS7grTO0nEMgVQ)\n\n## [引用我们](.\u002FCITATION)\n\n## 许可证\n\n版权所有 © 微软公司。保留所有权利。\n\n根据 [MIT](.\u002FLICENSE) 许可证授权。","# MARO 快速上手指南\n\nMARO (Multi-Agent Resource Optimization) 是微软开源的多智能体资源优化平台，旨在为现实世界的资源优化问题提供“强化学习即服务”（RaaS）。它广泛应用于物流（集装箱库存管理）、交通（共享单车调度）、数据中心（虚拟机调度）及金融资产管理等领域。\n\n## 1. 环境准备\n\n在开始之前，请确保您的系统满足以下要求：\n\n*   **操作系统**: Linux, macOS 或 Windows\n*   **Python 版本**: 支持 Python 3.x (具体版本请参考 PyPI 文件列表)\n*   **编译器依赖** (仅当从源码安装以使用 CLI 和可视化工具时需要):\n    *   **Linux \u002F macOS**: `gcc`\n    *   **Windows**: [Visual Studio 2017 Build Tools](https:\u002F\u002Fvisualstudio.microsoft.com\u002Fthank-you-downloading-visual-studio\u002F?sku=BuildTools&rel=15)\n\n> **注意**: 如果仅需使用 Python SDK 进行算法开发，可直接通过 pip 安装，无需编译器。若需使用命令行工具（如可视化仪表盘），建议从源码安装。\n\n## 2. 安装步骤\n\n### 方式一：通过 PyPI 安装（推荐用于快速开发）\n\n此方式安装核心 Python 包 `pymaro`，不包含 CLI 命令和可视化工具。\n\n**macOS \u002F Linux:**\n```sh\npip install pymaro\n```\n\n**Windows:**\nWindows 用户需先安装特定版本的 PyTorch，再安装 MARO：\n```powershell\n# 1. 安装 PyTorch (如未安装)\npip install torch===1.6.0 torchvision===0.7.0 -f https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Ftorch_stable.html\n\n# 2. 安装 MARO\npip install pymaro\n```\n\n### 方式二：从源码安装（包含 CLI 和可视化工具）\n\n如需使用 `maro` 命令行工具及可视化功能，请按以下步骤操作：\n\n**1. 创建并激活虚拟环境**\n\n*macOS \u002F Linux:*\n```sh\npython -m venv maro_venv\nsource .\u002Fmaro_venv\u002Fbin\u002Factivate\n```\n\n*Windows (PowerShell):*\n```powershell\npython -m venv maro_venv\n# 若遇到执行策略错误，运行以下命令解锁\nSet-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy Unrestricted\n.\\maro_venv\\Scripts\\activate\n```\n\n**2. 克隆代码并安装**\n\n```sh\n# 克隆仓库\ngit clone https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro.git\ncd maro\n```\n\n*macOS \u002F Linux:*\n```sh\nbash scripts\u002Finstall_maro.sh\npip install -r .\u002Frequirements.dev.txt\n```\n\n*Windows:*\n```powershell\n.\\scripts\\install_maro.bat\npip install -r .\u002Frequirements.dev.txt\n```\n\n**3. 配置环境变量**\n\n如果导入包时提示找不到模块，请设置 `PYTHONPATH`：\n\n*macOS \u002F Linux:*\n```sh\nexport PYTHONPATH=\u002Fpath\u002Fto\u002Fmaro\n```\n\n*Windows:*\n```powershell\n$Env:PYTHONPATH=\"C:\\path\\to\\maro\"\n```\n\n## 3. 基本使用\n\n以下是一个最简单的示例，展示如何初始化一个集装箱库存管理（CIM）场景并运行仿真。\n\n### 代码示例\n\n```python\nfrom maro.simulator import Env\n\n# 初始化环境\n# scenario: 场景名称 (cim = 集装箱库存管理)\n# topology: 拓扑结构\n# start_tick: 开始时间步\n# durations: 持续时长\nenv = Env(scenario=\"cim\", topology=\"toy.5p_ssddd_l0.0\", start_tick=0, durations=100)\n\n# 执行第一步\nmetrics, decision_event, is_done = env.step(None)\n\n# 运行直到结束\nwhile not is_done:\n    metrics, decision_event, is_done = env.step(None)\n\n# 输出最终指标\nprint(f\"environment metrics: {env.metrics}\")\n```\n\n### 可视化体验\n\n若已从源码安装，可开启数据快照并使用仪表盘查看仿真过程：\n\n1.  **启用数据导出**：在初始化 `Env` 时添加 `options` 参数。\n    ```python\n    env = Env(scenario=\"cim\",\n              topology=\"toy.5p_ssddd_l0.0\",\n              start_tick=0,\n              durations=100,\n              options={\"enable-dump-snapshot\": \".\u002Fdump_data\"})\n    # ... 运行 env.step() 逻辑 ...\n    ```\n\n2.  **启动可视化仪表盘**：\n    ```sh\n    maro inspector dashboard --source_path .\u002Fdump_data\u002FYOUR_SNAPSHOT_DUMP_FOLDER\n    ```\n\n### 快速体验 Playground\n\n如果您希望立即体验完整环境（包含 Jupyter Lab 和 Redis 监控），可以使用 Docker：\n\n```sh\n# 拉取镜像\ndocker pull maro2020\u002Fplayground\n\n# 运行容器\n# 访问 http:\u002F\u002F127.0.0.1:40010 使用 Jupyter Lab\n# 访问 http:\u002F\u002F127.0.0.1:40009 使用 Redis Commander\ndocker run -p 40009:40009 -p 40010:40010 maro2020\u002Fplayground\n```","某大型物流企业在全球港口网络中面临集装箱动态调配难题，需实时决定数千个集装箱在不同港口间的调运以平衡供需。\n\n### 没有 maro 时\n- 依赖人工经验或静态规则制定调运计划，无法应对突发的港口拥堵或需求波动，导致局部严重缺箱或空箱积压。\n- 传统优化算法难以处理海量状态空间，计算耗时过长，无法在分钟级内生成次日的全局最优调度方案。\n- 缺乏统一的多智能体仿真环境，难以评估不同调度策略在复杂连锁反应下的长期收益，试错成本极高。\n- 各港口数据孤岛严重，调度系统无法协同学习，往往陷入“头痛医头”的局部最优陷阱，整体运营成本高企。\n\n### 使用 maro 后\n- 基于强化学习构建多智能体协作模型，自动感知全局供需变化，动态生成抗干扰能力极强的实时调运策略。\n- 利用 MARO 内置的高效仿真引擎，将原本数小时的规划计算压缩至秒级，迅速输出覆盖全网络的精准调度指令。\n- 在安全的沙盒环境中大规模预训练智能体，提前模拟极端天气或罢工等场景，显著降低实际部署后的决策风险。\n- 打通各港口数据壁垒，通过多智能体协同进化实现全局资源最优配置，大幅降低空箱调运率和闲置等待时间。\n\nMARO 将复杂的资源调度转化为可落地的强化学习服务，帮助物流企业从“被动响应”转型为“主动预测”，显著降本增效。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmicrosoft_maro_61d56d65.gif","microsoft","Microsoft","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fmicrosoft_4900709c.png","Open source projects and samples from Microsoft",null,"opensource@microsoft.com","OpenAtMicrosoft","https:\u002F\u002Fopensource.microsoft.com","https:\u002F\u002Fgithub.com\u002Fmicrosoft",[82,86,90,94,98,102,106,110,114,117],{"name":83,"color":84,"percentage":85},"Python","#3572A5",67.1,{"name":87,"color":88,"percentage":89},"Jupyter Notebook","#DA5B0B",23.1,{"name":91,"color":92,"percentage":93},"Cython","#fedf5b",4.4,{"name":95,"color":96,"percentage":97},"C++","#f34b7d",3.5,{"name":99,"color":100,"percentage":101},"HTML","#e34c26",1.2,{"name":103,"color":104,"percentage":105},"Jinja","#a52a22",0.3,{"name":107,"color":108,"percentage":109},"Shell","#89e051",0.2,{"name":111,"color":112,"percentage":113},"CSS","#663399",0.1,{"name":115,"color":116,"percentage":113},"Batchfile","#C1F12E",{"name":118,"color":119,"percentage":120},"Dockerfile","#384d54",0,912,157,"2026-04-10T08:59:17","MIT",4,"Linux, macOS, Windows","未说明",{"notes":129,"python":130,"dependencies":131},"1. 若需使用 CLI 命令（包括可视化工具），必须从源代码安装，PyPI 包不包含这些功能。2. Windows 用户需要从源码安装时，需预先安装 Visual Studio 2017 Build Tools (C++ 编译器)。3. 建议使用虚拟环境 (venv) 进行安装。4. 从源码安装后，可能需要手动设置 PYTHONPATH 环境变量指向 MARO 目录。5. 提供 Docker 镜像 (maro2020\u002Fplayground) 以便快速体验，包含 Redis Commander 和 Jupyter Lab。","3.6, 3.7, 3.8 (根据 PyPI badge 推断，Windows 安装示例指定了 torch 1.6.0)",[132,133],"torch==1.6.0 (Windows 必需)","torchvision==0.7.0 (Windows 必需)",[13,135,14],"其他",[137,138,139,140,141,142,143,144,145,146,147,64,148,149,150,151],"reinforcement-learning","multi-agent-reinforcement-learning","multi-agent","simulator","resource-optimization","operations-research","citi-bike","inventory-management","logistics","raas","rl-algorithms","agent","docker","finance","transportation","2026-03-27T02:49:30.150509","2026-04-11T08:10:52.601219",[155,160,165,170,175,179],{"id":156,"question_zh":157,"answer_zh":158,"source_url":159},29039,"运行可视化命令 `maro inspector dashboard` 时出现 `ModuleNotFoundError: No module named 'maro.cli.maro_real_time_vis.back_end'` 错误怎么办？","该错误通常是因为快照数据未正确生成。请确保在初始化环境时添加了 `options={\"enable-dump-snapshot\": \".\u002Fdump_data\"}` 参数，并显式调用 `env.reset()` 来启动可视化所需的数据转储过程。示例代码如下：\n\n```python\nfrom maro.simulator import Env\nfrom maro.simulator.scenarios.cim.common import Action\n\nenv = Env(scenario=\"cim\",\n          topology=\"toy.5p_ssddd_l0.0\",\n          start_tick=0,\n          durations=100,\n          options={\"enable-dump-snapshot\": \".\u002Fdump_data\"})\nenv.reset()  # 必须调用此方法以生成可视化数据\n\nmetrics, decision_event, is_done = env.step(None)\naction = Action(0, 0, 0, 0)\n\nwhile not is_done:\n    metrics, decision_event, is_done = env.step(action)\n```\n\n注意：如果是单轮次（single epoch）运行，`maro inspector` 可能会报内部周期可视化错误，但跨周期可视化应能正常工作。","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fissues\u002F317",{"id":161,"question_zh":162,"answer_zh":163,"source_url":164},29040,"执行 `maro inspector geo --start service` 时提示 `dist folder does not exist` 或前端构建失败如何解决？","这通常是由于本地 Docker 镜像过旧导致的。请尝试删除本地的 `maro2020\u002Fgeo_front_service` 镜像，然后从 Docker Hub 重新拉取最新版本的镜像。具体命令如下：\n\n1. 删除旧镜像：`docker rmi maro2020\u002Fgeo_front_service`\n2. 拉取新镜像：`docker pull maro2020\u002Fgeo_front_service`\n\n完成后再次运行启动命令即可。","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fissues\u002F314",{"id":166,"question_zh":167,"answer_zh":168,"source_url":169},29041,"在 CIM 场景中，如何获取船舶当前是否停靠以及停靠港口的索引信息？","项目已在主分支（master branch）中更新了船舶属性以支持此功能。您可以访问船舶对象的以下两个新属性：\n\n- `is_parking`: 表示船舶是否正在停靠（1 表示停靠，0 表示航行）。\n- `loc_port_idx`: 表示船舶当前停靠的港口索引。\n\n请确保您的代码基于最新的 master 分支版本。","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fissues\u002F360",{"id":171,"question_zh":172,"answer_zh":173,"source_url":174},29042,"在 macOS 上从源码编译 MARO 时遇到 `clang: error: no such file or directory: '.\u002Fmaro\u002Fbackends\u002Fbackend.c'` 错误怎么办？","该错误通常是因为构建脚本未正确生成 C 后端文件。请尝试先清理之前的构建缓存，然后重新运行构建命令。如果问题依旧，请确保您使用的是项目最新版本的代码，因为该问题可能在后续提交中已修复。确认修复后，再次运行 `python setup.py build` 即可成功编译。","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fissues\u002F207",{"id":176,"question_zh":177,"answer_zh":178,"source_url":169},29043,"CIM 场景中 `VesselFutureStopsPrediction` 预测的未来停靠点列表出现偏差（提前一步）是什么原因？","这是一个已知的逻辑 Bug，当 `last_loc_idx == next_loc_idx` 时，预测结果会错误地向前跳跃一步。该问题已在后续的 PR（如 #366）中修复。如果您遇到此问题，请将 MARO 更新到最新的 master 分支版本，以获取修正后的预测逻辑。",{"id":180,"question_zh":181,"answer_zh":182,"source_url":183},29044,"为什么 `future_stop_tick_list` 中会出现小于当前时间步（current tick）的预测值？","正常情况下，未来停靠时间步列表中的值应始终大于当前时间步。如果出现小于当前时间步的值（例如当前为 100，列表中却有 75），这属于异常行为，通常是由内部状态同步或预测逻辑的 Bug 引起的。建议检查您使用的 MARO 版本，并升级到最新的 master 分支，因为开发团队已针对此类预测逻辑错误进行了多次修复和优化。","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fissues\u002F384",[185,190,195,200,205,210,215,220,225,230,235,240,245,250,255,260,265,270,275,280],{"id":186,"version":187,"summary_zh":188,"released_at":189},197890,"maro-0.3.2a4","- 添加双深度Q网络（DDQN）\n- 添加优先级采样\n- 改进探索策略\n- 修复 minor 问题","2023-10-27T06:21:18",{"id":191,"version":192,"summary_zh":193,"released_at":194},197891,"maro-0.3.2a3","- 修复 DiscreteRLPolicy 的 set_state 漏洞（https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fcommit\u002Fa9a638134bef84176a0623fdadc5b789c2d7c3cf，https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fcommit\u002Fb3c6a589ad9036b03221e776a6929b2bc1eb4680）","2023-05-15T04:34:19",{"id":196,"version":197,"summary_zh":198,"released_at":199},197892,"maro-0.3.2a1","- 优化强化学习工作流\n  + 在函数中添加 `**kwargs`，以支持更多问题设置（例如基于图的问题）(#589)\n     - 将 `**kwargs` 添加到强化学习模型的 `forward` 方法和 `_shape_check()` 中\n     - 将 `**kwargs` 添加到强化学习策略的 `get_action` 相关方法和 `_post_check()` 中\n     - 将 `**kwargs` 添加到 `AbsEnvSampler` 的 `choose_actions` 方法中；在当前的 `sample()` 和 `eval()` 方法中仍保持为 `None`\n  + 在当前 `TrainOps` 的 `update_critic()` 和 `update_actor()` 返回值中加入 detached loss；同时为当前 `TrainOps` 的 `update_actor()` 方法添加默认值为 `False` 的 `early_stop` 参数 (#589)\n  + 优化强化学习工作流中的随机种子设置逻辑 (#584)\n  + 优化 rollout 工作流 (#577)，以支持：\n    - 在 rollout 过程中运行指定步数\n    - 在评估过程中使用 `num_eval_episodes` 运行指定数量的 episode\n    - 使用 `AbsEnvSampler.metrics` 在 rollout 过程中灵活管理指标\n  + 添加 `AbsEnvSampler.metrics`，以支持在 rollout 过程中灵活管理指标 (#577)\n  + 添加 `Callback` 作为通用接口，用于支持工作流各阶段的自定义操作。\n       - 新增两个实例 `Checkpoint` 和 `MetricsRecorder`。(#577)\n       - 将 `customized_callbacks` 添加到 `RLComponentBundle` 中。(#589)\n  + 重新组织强化学习任务的输出路径。(#577)\n  + 修复若干强化学习算法中的 bug。(#577, #589)\n- 在整个项目中将 NumPy 数据类型替换为 Python 常用数据类型 (#571)\n- 在 *tests\u002F* 目录下新增 Mujoco 上的强化学习基准测试模块，与 spinning up 基准测试相比，性能结果可参见 *tests\u002Frl\u002Fperformance.md* (#575, #577, #583, #584)\n- 其他代码的小幅优化","2023-03-30T02:01:33",{"id":201,"version":202,"summary_zh":203,"released_at":204},197893,"maro-0.3.1a2","更新构建工作流，以修复 pymaro 包在 manylinux 环境下的失败问题。","2023-03-15T02:16:17",{"id":206,"version":207,"summary_zh":208,"released_at":209},197894,"maro-0.3.1a1","- 优化决策事件逻辑 (#559)\n  - 添加 `DecisionEventPayload` 和 `ActionPayload` 作为决策事件和动作事件负载的基类。\n  - 在运行时添加相关类型检查。\n  - 重命名 `simulator\u002F` 目录下的相关变量。\n  - 优化动作处理逻辑，修改 `Env` 及相关后端组件。\n- 优化强化学习组件包逻辑 (#549)\n  - 优化 `rl_component_bundle`，采用更直观、更易用的方式进行组织。在新版本中，不再依赖 easyrl。\n  - 更新相关示例。\n  - 修复分布式训练中的 bug，并为 CIM 场景的分布式训练添加配置 YAML 文件。\n  - 新增 `rl_formulation.ipynb` 作为示例。\n- 修复 CITI BIKE 世界你好仪表板 (#555)\n- 更新数据模型文档 (#554)\n- 更新需求文档 (#552, #553)","2022-12-27T09:55:18",{"id":211,"version":212,"summary_zh":213,"released_at":214},197895,"maro-0.3.0a1","- 强化学习工具包：✨ 强化学习工具包全新设计 (#539)  \n- 命令行界面：重构 (#539)  \n- 后端：修复数据帧\u002F快照精度问题 (#544)  \n- 其他：整个仓库代码格式化 (#538, #547)  \n- 摩拜单车可视化工具紧急修复 (#543)","2022-06-14T08:09:43",{"id":216,"version":217,"summary_zh":218,"released_at":219},197896,"maro-0.2.4a1","- 添加更高版本的 Python 3.8、3.9 (#398)\n- 对部分核心模块进行优化，包括增加类型注解、改进代码风格、小幅接口调整以及潜在 bug 修复\n  - 核心及抽象业务引擎优化 (#392)\n  - 事件缓冲优化 (#389)\n  - SimRandom 接口更新，以更好地支持场景构建 (#401, #400)\n- CIM 场景优化 (#400)\n  - 禁用自动动作类型检测，即需显式指定 `ActionType` (#399)\n  - 数据处理更新，涵盖数据生成部分和真实数据加载部分 (#395)\n  - 部分测试程序更新，并扩大测试覆盖范围 (#395)\n- 示例更新\n  - Citi Bike 的 OnlineLP：将默认求解器由 GLPK 更改为 CBC (#391)","2021-09-27T02:59:09",{"id":221,"version":222,"summary_zh":223,"released_at":224},197897,"maro-0.2.3a4","场景\u002FCIM：\n- 修复在调用 Env.reset() 后，针对某些特定拓扑结构的船舶规划可能出现的 bug（相关 issue：#385，相关 PR：#387、#388）。","2021-08-26T06:51:07",{"id":226,"version":227,"summary_zh":228,"released_at":229},197898,"maro-0.2.3a3","模拟器\u002FCIM 场景：\n- 修复环境重置问题 #385，并更新模拟器随机种子接口 (#387)\n- 修复“未来停止点预测错误”问题 #384 (#386)\n\n强化学习工具包：\n- 将 `Actor` 使用的参数添加到 `Trajectory.on_env_feedback()` 接口中 (#373, #374)\n\n可视化工具\u002FCIM：\n- 修复 Geo 可视化 IP 地址及 SQL 逻辑中的 bug (#352, #383)\n\n其他：\n- 更新 setup 中的 dataclasses 依赖项\n- 修复代码注释和在线文档中的一些拼写错误\n","2021-08-19T07:52:20",{"id":231,"version":232,"summary_zh":233,"released_at":234},197899,"maro-0.2.3a2","* 实时数据模式下的 CIM 数据容器\u002F数据加载器接口已更新 (#372)\n* pymaro 说明的 README 已更新","2021-07-13T06:16:43",{"id":236,"version":237,"summary_zh":238,"released_at":239},197900,"maro-0.2.3a1","CIM Scenario:\r\n* Add the data container and data loader for real data schema (#361)\r\n* Fix the vessel plan, vessel past\u002Ffuture stops info issues (#363) \r\n* Add attributes `is_parking` and `loc_port_idx` for vessels (#366)\r\n\r\nSimulator:\r\n* Add support to dump finished events (#365)\r\n* Add property `business_engine` to `AbsEnv` (#368)\r\n\r\nRL Toolkit:\r\n* Fixed scheduler init bug (#362)\r\n\r\nTutorials docker images:\r\n* Fix bugs in examples and notebooks caused by interface changes. (#367)\r\n* Update simple strategies used in examples (E.g. random instead of dumpy in CIM hello series). (#367)\r\n* Update docker images and related documents. (#367)","2021-07-09T07:25:44",{"id":241,"version":242,"summary_zh":243,"released_at":244},197901,"maro-0.2.2a3","Package dependency:\r\n- Update package dependency (#345 #346)\r\n\r\nSimulator:\r\n- Fixed the vessel plan issue for CIM scenario (#348)\r\n\r\nExamples:\r\n- Update hello world examples for CIM scenario (#338)","2021-05-31T03:15:46",{"id":246,"version":247,"summary_zh":248,"released_at":249},197902,"maro-0.2.2a2","Package dependency:\r\n- Loosen package dependency (#328) to be more flexible: #331\r\n\r\nCLI related:\r\n- Fixed issues (#317, #314) for geographic visualization tool: #325 \r\n- Fixed Redis database data not cleaned issue (#327) by adding a close function for Proxy: #335 \r\n\r\nExamples:\r\n- Add offline ILP example for VM scheduling scenario: #275 ","2021-05-14T03:35:42",{"id":251,"version":252,"summary_zh":253,"released_at":254},197903,"maro-0.2.2a1","Backend related:\r\n- bug fix: dynamic backend slot number not correct after reset (#321)\r\n- bug fix: value is interrupt if one node contains multiple list attribute (#321)\r\n- changes: snapshot query result default value set to 0 instead nan, make it align with static backend (#321)\r\n- bug fix: on windows compiling fail due to cannot find max function. (#321)\r\n\r\nRL Toolkit related:\r\n- Fix bug of dqn.choose_action and ddpg.choose_action (#307)\r\n- Double DQN: loss function bug fixed (#315)\r\n- Fix bug caused by invalid snapshot index (#318, #319)\r\n\r\nCLI related:\r\n- Add MARO admin tool (#288)\r\n- Add master\u002Fnode vm join deployment file for grass\u002Fon-premises. (#316)\r\n- Visualization tool: Add support of dump data path creation, default sampling ratio to 1 (#323)\r\n- Visualization tool: Bug occurs in Linux fixed (#312, #313)\r\n- Update online doc Installation part (#316); Visualization part (#301)","2021-04-15T08:36:12",{"id":256,"version":257,"summary_zh":258,"released_at":259},197904,"maro-0.2.1a1","- Backend related:\r\n  - Vector env support (#266)\r\n  - Add slot filter functions for node attribute (#273)\r\n  - Refine joint decision sequential action mode (#219)\r\n  - Add dynamic node support (#172)\r\n- VM scheduling scenario related:\r\n  - Add oversubscription feature (#246), (#256)\r\n  - Add hierarchy vm region architecture support (#258)\r\n  - Fix bug of auto-downloading data (#257)\r\n  - Pricing model and energy model update (#286)\r\n  - Rule-based algorithm examples (#255), (#282)\r\n- RL toolkit related:\r\n  - Add DDPG algorithm (#252)\r\n  - Merge algorithm with agent (#259)\r\n  - Distributed framework update (#206)\r\n- CLI related:\r\n  - CLI refactoring (#227)\r\n  - Feature: Add a cli command to support create new project (#279)\r\n  - Add Env-Geographic visualization tool, CIM hello as example (#291), (#294), (#295), (#298)\r\n  - CLI visualization support and maro grass local mode (#277) ","2021-03-22T07:17:12",{"id":261,"version":262,"summary_zh":263,"released_at":264},197905,"maro-0.2.0a1","# Description\r\n\r\n- Simulation toolkit\r\n  - VM allocation environment\r\n  - Dashboard for CIM and CitiBike\r\n  - EventBuffer refine\r\n  - Dump for snapshots, events and business engine special data.\r\n- RL toolkit\r\n  - New RL abstraction\r\n  - More RL algorithms \r\n- Distributed toolkit\r\n  - maro cli refactoring\r\n  - maro cli support windows \r\n  - maro processes mode\r\n  - maro grass on-premises mode\r\n- Solution example\r\n  - CIM: dqn & gnn example updated\r\n  - Citi Bike: online LP example added\r\n  - VM Scheduling: random & best fit example added\r\n\r\n## Linked issue(s)\u002FPull request(s)\r\n\r\n\u003C!--Please add the related issue link(s) below.-->\r\n- [MARO processes mode doc](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F230)\r\n- [MARO on-premised mode](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F220)\r\n- [Learning model refactoring](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F236)\r\n- [MARO vis](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F233)\r\n- [MARO vis dump feature](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F190)\r\n- [RL toolkit doc update](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F235)\r\n- [VM scheduling doc](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F228)\r\n- [VM scheduling environment](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F189)\r\n- [VM data pipeline](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F199)\r\n- [EventBuffer refine](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F197)","2021-01-04T11:53:05",{"id":266,"version":267,"summary_zh":268,"released_at":269},197906,"maro-0.1.2a2","# Description\r\n\r\nThis PR contains several PRs including some bug fix, new example addition, doc addition and refinement, new features, new tests.\r\n\r\n## Linked issue(s)\u002FPull request(s)\r\n\r\n- [138](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F138):\r\n  - refine data push\u002Fpull in cli\r\n- [112](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F112):\r\n  - add fall back function in weather download\r\n- [136](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F136):\r\n  - add docs for example\r\n- [144](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F144):\r\n  - switching the key and values of the handler dict which is used in the dist decorator\r\n  - add unit tests for the dist decorator\r\n  - fixed the multithreading deadlock in the maro unit test suite\r\n- [147](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F147):\r\n  - remove useless reward() from Env, BE, yaml config\r\n  - refine the format & style for core & BE\r\n- [157](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F157):\r\n  - add docs for AzCopy\r\n  - add dqn tests in grass\u002Fk8s mode\r\n- [160](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F160), [164](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F164):\r\n  - fix the data generation description for citi bike in docs\r\n- [60](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F60):\r\n  - add example\u002Fcim\u002Fgnn\r\n- [166](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F166):\r\n  - fix the uppicklable bug in RL toolkit store\r\n- [167](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F167):\r\n  - rename the confusing components in RL toolkit and example\u002Fcim\u002Fdqn","2020-11-05T03:24:42",{"id":271,"version":272,"summary_zh":273,"released_at":274},197907,"maro-0.1.1a11","- Fix dist decorator\r\n- Add example docs for CIM and Citi Bike scenario\r\n- Refine annotation\r\n\r\n[PR 153](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fmaro\u002Fpull\u002F153)","2020-10-13T05:52:20",{"id":276,"version":277,"summary_zh":278,"released_at":279},197908,"maro-0.1.1a10","- Add fallback logic for weather source unavailable case.\r\n- Refine CLI data pull\u002Fpush logic.\r\n- Add integration test cases for orchestration.","2020-10-08T11:56:05",{"id":281,"version":282,"summary_zh":283,"released_at":284},197909,"maro-0.1.1a9","Refine docstring.","2020-09-30T04:56:21"]