[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-facebookexperimental--Robyn":3,"tool-facebookexperimental--Robyn":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",160015,2,"2026-04-18T11:30:52",[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。",109154,"2026-04-18T11:18:24",[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":76,"owner_twitter":76,"owner_website":77,"owner_url":78,"languages":79,"stars":105,"forks":106,"last_commit_at":107,"license":108,"difficulty_score":109,"env_os":110,"env_gpu":111,"env_ram":111,"env_deps":112,"category_tags":120,"github_topics":122,"view_count":32,"oss_zip_url":76,"oss_zip_packed_at":76,"status":17,"created_at":135,"updated_at":136,"faqs":137,"releases":167},9258,"facebookexperimental\u002FRobyn","Robyn","Robyn is an experimental, AI\u002FML-powered and open sourced Marketing Mix Modeling (MMM) package from Meta Marketing Science. Our mission is to democratise modeling knowledge, inspire the industry through innovation, reduce human bias in the modeling process & build a strong open source marketing science community.","Robyn 是由 Meta Marketing Science 推出的开源营销组合模型（MMM）工具，旨在利用人工智能与机器学习技术，让复杂的营销效果评估变得自动化且易于获取。它主要解决了传统建模方法成本高、依赖人工经验且难以适应隐私保护趋势的痛点，帮助企业在无法追踪个体用户数据的情况下，依然能科学地衡量各媒体渠道的真实贡献。\n\n这款工具特别适合拥有丰富数据源的数字广告主、直接响应型营销人员，以及希望深入探索营销科学的数据分析师和研究人员。无论是大型品牌还是中小型企业，都能通过 Robyn 降低建模门槛，优化预算分配。\n\n在技术层面，Robyn 展现了强大的自动化能力。它融合了岭回归、多目标进化算法（用于超参数优化）、时间序列分解（识别趋势与季节性）以及基于梯度的预算优化等先进算法。这些技术不仅能自动探索广告衰减率和饱和曲线，还能有效减少人为偏见，提供更为客观的渠道效率分析。目前 Robyn 主要支持 R 语言，并提供了实验性的 Python 版本，配合详细的演示脚本，帮助用户快速上手并构建属于自己的营销评估模型。","# Robyn: Continuous & Semi-Automated MMM \u003Cimg src='https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_91e1ce1df128.png' align=\"right\" height=\"139px\" \u002F>\n### The Open Source Marketing Mix Model Package from Meta Marketing Science\n\n[![CRAN\\_Status\\_Badge](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_5def36762984.png)](https:\u002F\u002Fcran.r-project.org\u002Fpackage=Robyn) [![Downloads](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_1e8edb101ae8.png)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_1e8edb101ae8.png) [![Site](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fsite-Robyn-blue.svg)](https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002F) [![Facebook](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fgroup-Facebook-blue.svg)](https:\u002F\u002Fwww.facebook.com\u002Fgroups\u002Frobynmmm\u002F) [![CodeFactor](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_05765eef716e.png)](https:\u002F\u002Fwww.codefactor.io\u002Frepository\u002Fgithub\u002Ffacebookexperimental\u002Frobyn)\n---\n\n## Introduction\n\n  * **What is Robyn?**: Robyn is an experimental, semi-automated and open-sourced Marketing Mix Modeling (MMM) package from Meta Marketing Science. It uses various machine learning techniques (Ridge regression, multi-objective evolutionary algorithm for hyperparameter optimization, time-series decomposition for trend & season, gradient-based optimization for budget allocation, clustering, etc.) to define media channel efficiency and effectivity, explore adstock rates and saturation curves. It's built for granular datasets with many independent variables and therefore especially suitable for digital and direct response advertisers with rich data sources. \n  \n  * **Why are we doing this?**: MMM used to be a resource-intensive technique that was only affordable for \"big players\". As the privacy needs of the measurement landscape evolve, there's a clear trend of increasing demand for modern MMM as a privacy-safe solution. At Meta Marketing Science, our mission is to help all businesses grow by transforming marketing practices grounded in data and science. It's highly aligned with our mission to democratizing MMM and making it accessible for advertisers of all sizes. With Project Robyn, we want to contribute to the measurement landscape, inspire the industry and build a community for exchange and innovation around the future of MMM and Marketing Science in general.\n  \nRobyn is available in R and Python. For installation and usage guide see below. Please note that the current Python version is a LLM-translated Beta version and might encounter bugs. \n  \n## Quick start for R\n\n**1. Installing the package**\n  \n  * Install Robyn latest R package version:\n```{r}\n## CRAN VERSION\ninstall.packages(\"Robyn\")\n\n## DEV VERSION\n# If you don't have remotes installed yet, first run: install.packages(\"remotes\")\nremotes::install_github(\"facebookexperimental\u002FRobyn\u002FR\")\n```\n\n  * If it's taking too long to download, you have a slow or unstable internet connection, and have [issues](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fissues\u002F309) while installing the package, try setting `options(timeout=400)`.\n  \n  * Robyn requires the Python library [Nevergrad](https:\u002F\u002Ffacebookresearch.github.io\u002Fnevergrad\u002F). If encountering Python-related \n  error during installation, please check out the [step-by-step guide](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fblob\u002Fmain\u002Fdemo\u002Fdemo.R) as well as this [issue](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fissues\u002F189) to get more info.\n  \n  * For Windows, if you get openssl error, please see instructions\n  [here](https:\u002F\u002Fstackoverflow.com\u002Fquestions\u002F54558389\u002Fhow-to-solve-this-error-while-installing-python-packages-in-rstudio\u002F54566647) and\n  [here](https:\u002F\u002Fdev.to\u002Fdanilovieira\u002Finstalling-openssl-on-windows-and-adding-to-path-3mbf) to install and update openssl.\n\n**2. Getting started**\n\n  * Use this [demo.R](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Ftree\u002Fmain\u002Fdemo\u002Fdemo.R) script as step-by-step guide that is\n  intended to cover most common use-cases. Test the package using simulated dataset provided in the package. \n  \n  * Visit our [website](https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002F) to explore more details about Project Robyn.\n  \n  * Join our [public group](https:\u002F\u002Fwww.facebook.com\u002Fgroups\u002Frobynmmm\u002F) to exchange with other users and interact with team Robyn.\n  \n  * Take Meta's [official Robyn blueprint course](https:\u002F\u002Fwww.facebookblueprint.com\u002Fstudent\u002Fpath\u002F253121-marketing-mix-models?utm_source=readme) online \n  \n## Quick start for Python (Beta)\n\nThe Python version of Robyn is rewritten from Robyn's R package version `3.11.1` to Python using object oriented programming principles and modular architecture for a robust solution. It was developed by utilizing various LLMs and AI workflows like [Llama](https:\u002F\u002Fwww.llama.com\u002F). As is common with any AI-based solutions, there may be potential challenges in translating code from one language to another. In this case, we anticipate that there could be some issues in the translation from R to Python. However, we believe in the power of community collaboration and open-source contribution. Therefore, we are opening this project to the community to participate and contribute. Together, we can address and resolve any issues that may arise, enhancing the functionality and efficiency of the Python version of Robyn. We look forward to your contributions and to the continuous improvement of this project.\n\n### 1. Prerequisites\n\n- R must be installed on your machine. Download from the [official R Project website](https:\u002F\u002Fwww.r-project.org\u002F)\n- The glmnet R package is required\n\n#### Installing glmnet on Windows\n```bash\n# Open R console (run in Command Prompt\u002FPowerShell)\nR\n# Then in R console, install glmnet\ninstall.packages(\"glmnet\")\n# Exit R console\nq()\n```\n\n#### Installing glmnet on macOS\u002FLinux\n```bash\n# Open terminal and run R\nR\n# Then in R console, install glmnet\ninstall.packages(\"glmnet\")\n# Exit R console\nq()\n```\n\n### 2. Setting up Python Environment\n\n#### Windows\n```bash\n# Create virtual environment\npython -m venv robyn-env\n\n# Activate virtual environment\nrobyn-env\\Scripts\\activate\n```\n\n#### macOS\u002FLinux\n```bash\n# Create virtual environment\npython3 -m venv robyn-env\n\n# Activate virtual environment\nsource robyn-env\u002Fbin\u002Factivate\n```\n\n### 3. Install Robyn\n\nChoose one of the following installation methods:\n\n```bash\n# Install from PyPI (recommended)\npip3 install robynpy\n\n# OR install development version from source\npip3 install -r requirements.txt\n```\n\n### 4. Getting Started\n\nThe `python\u002Fsrc\u002Frobyn\u002Ftutorials` directory contains tutorials for most common scenarios using the included simulated dataset.\n\nThere are two ways to run Python Robyn:\n\n#### Option 1: Using tutorial1.ipynb (Recommended)\n- Provides an end-to-end flow with a one-click solution\n- Ideal for users who prefer minimal setup\n- Uses APIs from `python\u002Fsrc\u002Frobyn\u002Frobyn.py`\n- Includes feature engineering, model training, clustering, one-pager generation, and budget allocation\n- Configurations can be modified directly in the notebook\n\n#### Option 2: Using tutorial1_src.ipynb (Advanced)\n- Offers more flexibility and control over individual modules\n- Designed for users who want to customize the workflow\n- Calls modules directly with parameters\n- Allows skipping specific components (clustering\u002Fone-pager plots\u002Fbudget allocation)\n- Requires understanding of underlying logic\n  \n## Quick start Python wrapper (Robyn API for Python beta)\n\nThe Robyn API for Python (beta), first released on Nov.22nd 2023, is a plumber-based solution that requires the installation of the Robyn R pacakge first. It serves as a workaround when the Python native version is not yet available or up-to-date. Please see the usage guide [here](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fblob\u002Fmain\u002Frobyn_api\u002Frobyn_python_notebook.ipynb).\n\n\n## License\n\nMeta's Robyn is MIT licensed, as found in the LICENSE file.\n\n- Terms of Use - https:\u002F\u002Fopensource.facebook.com\u002Flegal\u002Fterms \n- Privacy Policy - https:\u002F\u002Fopensource.facebook.com\u002Flegal\u002Fprivacy\n- Defensive Publication - https:\u002F\u002Fwww.tdcommons.org\u002Fdpubs_series\u002F4627\u002F\n\n## Contact\n\n* gufeng@meta.com, Gufeng Zhou, Marketing Science, Robyn creator\n* igorskokan@meta.com, Igor Skokan, Marketing Science Director, open source\n","# Robyn：持续且半自动化的 MMM \u003Cimg src='https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_91e1ce1df128.png' align=\"right\" height=\"139px\" \u002F>\n### 来自 Meta Marketing Science 的开源营销组合模型工具包\n\n[![CRAN\\_Status\\_Badge](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_5def36762984.png)](https:\u002F\u002Fcran.r-project.org\u002Fpackage=Robyn) [![Downloads](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_1e8edb101ae8.png)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_1e8edb101ae8.png) [![Site](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fsite-Robyn-blue.svg)](https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002F) [![Facebook](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fgroup-Facebook-blue.svg)](https:\u002F\u002Fwww.facebook.com\u002Fgroups\u002Frobynmmm\u002F) [![CodeFactor](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_readme_05765eef716e.png)](https:\u002F\u002Fwww.codefactor.io\u002Frepository\u002Fgithub\u002Ffacebookexperimental\u002Frobyn)\n---\n\n## 简介\n\n  * **什么是 Robyn？**：Robyn 是 Meta Marketing Science 推出的一款实验性、半自动化且开源的营销组合建模（MMM）工具包。它利用多种机器学习技术（岭回归、用于超参数优化的多目标进化算法、用于趋势与季节性分解的时间序列分析、基于梯度的预算分配优化、聚类等），来定义媒体渠道的效率和效果，并探索广告库存率及饱和曲线。该工具包专为包含大量自变量的细粒度数据集而设计，因此特别适合拥有丰富数据源的数字广告和直接响应型广告主。\n  \n  * **我们为何要开发它？**：过去，MMM 是一种资源密集型技术，只有大型企业才能负担得起。随着测量领域对隐私保护需求的不断演变，现代 MMM 作为一种隐私友好的解决方案，其市场需求正日益增长。在 Meta Marketing Science，我们的使命是通过以数据和科学为基础的营销实践变革，帮助所有企业实现增长。这与我们致力于 democratizing MMM、使其为各类规模的广告主所用的愿景高度一致。借助 Project Robyn，我们希望为测量行业贡献力量，激发业界创新，并围绕 MMM 及整个营销科学的未来建立交流与创新的社区。\n  \nRobyn 同时提供 R 和 Python 版本。安装及使用指南见下文。请注意，当前的 Python 版本是由大语言模型翻译的测试版，可能存在一些 bug。\n  \n## R 快速入门\n\n**1. 安装软件包**\n  \n  * 安装最新版本的 Robyn R 包：\n```{r}\n## CRAN 版本\ninstall.packages(\"Robyn\")\n\n## 开发版本\n# 如果尚未安装 remotes 包，请先运行：install.packages(\"remotes\")\nremotes::install_github(\"facebookexperimental\u002FRobyn\u002FR\")\n```\n\n  * 如果下载时间过长，可能是因为您的网络连接较慢或不稳定。若在安装过程中遇到问题（参见 [GitHub 问题 #309](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fissues\u002F309)），可尝试设置 `options(timeout=400)`。\n  \n  * Robyn 需要 Python 库 [Nevergrad](https:\u002F\u002Ffacebookresearch.github.io\u002Fnevergrad\u002F)。如果在安装过程中遇到与 Python 相关的错误，请参考 [逐步指南](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fblob\u002Fmain\u002Fdemo\u002Fdemo.R) 以及此 [问题](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fissues\u002F189)，以获取更多信息。\n  \n  * 对于 Windows 用户，若出现 openssl 错误，请参阅以下说明：\n  [这里](https:\u002F\u002Fstackoverflow.com\u002Fquestions\u002F54558389\u002Fhow-to-solve-this-error-while-installing-python-packages-in-rstudio\u002F54566647) 和\n  [这里](https:\u002F\u002Fdev.to\u002Fdanilovieira\u002Finstalling-openssl-on-windows-and-adding-to-path-3mbf) 提供了安装和更新 openssl 的步骤。\n\n**2. 开始使用**\n\n  * 使用此 [demo.R](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Ftree\u002Fmain\u002Fdemo\u002Fdemo.R) 脚本作为分步指南，旨在覆盖大多数常见用例。请使用软件包中提供的模拟数据集测试该工具包。\n  \n  * 访问我们的 [网站](https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002F)，了解更多关于 Project Robyn 的详细信息。\n  \n  * 加入我们的 [公开群组](https:\u002F\u002Fwww.facebook.com\u002Fgroups\u002Frobynmmm\u002F)，与其他用户交流并直接与 Robyn 团队互动。\n  \n  * 在线学习 Meta 的 [官方 Robyn 蓝图课程](https:\u002F\u002Fwww.facebookblueprint.com\u002Fstudent\u002Fpath\u002F253121-marketing-mix-models?utm_source=readme)。\n\n## Python 快速入门（Beta）\n\nRobyn 的 Python 版本基于 Robyn 的 R 包版本 `3.11.1`，采用面向对象编程原则和模块化架构重新编写，以构建一个稳健的解决方案。该版本由多种大语言模型和 AI 工作流（如 Llama）协助开发而成。如同任何基于 AI 的解决方案一样，在将代码从一种语言翻译到另一种语言的过程中，可能会遇到一些挑战。在此案例中，我们预计从 R 到 Python 的翻译过程中可能会出现一些问题。然而，我们坚信社区协作与开源贡献的力量。因此，我们向社区开放此项目，邀请大家参与并贡献力量。通过共同努力，我们可以解决可能出现的问题，从而提升 Robyn Python 版本的功能性和效率。我们期待您的参与，也期待该项目的持续改进。\n\n### 1. 先决条件\n\n- 您的计算机上必须安装 R。请从 [R 官方网站](https:\u002F\u002Fwww.r-project.org\u002F) 下载。\n- 需要安装 glmnet R 包。\n\n#### 在 Windows 上安装 glmnet\n```bash\n# 打开 R 控制台（在命令提示符或 PowerShell 中运行）\nR\n# 然后在 R 控制台中安装 glmnet\ninstall.packages(\"glmnet\")\n# 退出 R 控制台\nq()\n```\n\n#### 在 macOS\u002FLinux 上安装 glmnet\n```bash\n# 打开终端并运行 R\nR\n# 然后在 R 控制台中安装 glmnet\ninstall.packages(\"glmnet\")\n# 退出 R 控制台\nq()\n```\n\n### 2. 设置 Python 环境\n\n#### Windows\n```bash\n# 创建虚拟环境\npython -m venv robyn-env\n\n# 激活虚拟环境\nrobyn-env\\Scripts\\activate\n```\n\n#### macOS\u002FLinux\n```bash\n# 创建虚拟环境\npython3 -m venv robyn-env\n\n# 激活虚拟环境\nsource robyn-env\u002Fbin\u002Factivate\n```\n\n### 3. 安装 Robyn\n\n请选择以下任一安装方式：\n\n```bash\n# 从 PyPI 安装（推荐）\npip3 install robynpy\n\n# 或者从源码安装开发版本\npip3 install -r requirements.txt\n```\n\n### 4. 开始使用\n\n`python\u002Fsrc\u002Frobyn\u002Ftutorials` 目录包含使用附带的模拟数据集的常见场景教程。\n\n运行 Python Robyn 有两种方式：\n\n#### 选项 1：使用 tutorial1.ipynb（推荐）\n- 提供端到端的工作流，一键式解决方案\n- 非常适合偏好极简设置的用户\n- 使用 `python\u002Fsrc\u002Frobyn\u002Frobyn.py` 中的 API\n- 包括特征工程、模型训练、聚类、生成概览报告以及预算分配\n- 可以直接在笔记本中修改配置\n\n#### 选项 2：使用 tutorial1_src.ipynb（进阶）\n- 提供更高的灵活性和对各个模块的控制能力\n- 专为希望自定义工作流程的用户设计\n- 直接通过参数调用各个模块\n- 允许跳过特定组件（聚类\u002F概览图表\u002F预算分配）\n- 需要理解底层逻辑\n\n## Python 封装快速入门（Robyn Python API 测试版）\n\nRobyn Python API（测试版）于 2023 年 11 月 22 日首次发布，是一种基于管道的解决方案，需要先安装 Robyn R 包。当 Python 原生版本尚未可用或未更新时，它可以作为一种临时替代方案。请参阅使用指南 [此处](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fblob\u002Fmain\u002Frobyn_api\u002Frobyn_python_notebook.ipynb)。\n\n## 许可证\n\nMeta 的 Robyn 采用 MIT 许可证，具体信息见 LICENSE 文件。\n- 使用条款 - https:\u002F\u002Fopensource.facebook.com\u002Flegal\u002Fterms\n- 隐私政策 - https:\u002F\u002Fopensource.facebook.com\u002Flegal\u002Fprivacy\n- 防御性公开文档 - https:\u002F\u002Fwww.tdcommons.org\u002Fdpubs_series\u002F4627\u002F\n\n## 联系方式\n\n* gufeng@meta.com，Gufeng Zhou，营销科学部门，Robyn 创建者\n* igorskokan@meta.com，Igor Skokan，营销科学总监，开源项目负责人","# Robyn 快速上手指南\n\nRobyn 是由 Meta Marketing Science 开源的营销组合模型（MMM）工具，支持 R 和 Python。它利用机器学习技术（如岭回归、多目标进化算法等）自动分析媒体渠道效率、广告衰减率及饱和度曲线，特别适用于拥有丰富数据源的数字广告主。\n\n## 环境准备\n\n### 系统要求\n- **操作系统**：Windows \u002F macOS \u002F Linux\n- **核心依赖**：\n  - **R 语言**：必须安装（即使使用 Python 版本也需要 R 作为底层支撑）。\n  - **Python**：建议使用 Python 3.8+（用于运行 Python 原生版或 API 封装版）。\n\n### 前置依赖检查\n1. **安装 R**：前往 [CRAN](https:\u002F\u002Fcran.r-project.org\u002F) 下载并安装最新版 R。\n2. **安装 R 包 `glmnet`**（Python 版必需）：\n   打开 R 控制台执行：\n   ```r\n   install.packages(\"glmnet\")\n   ```\n3. **Python 依赖**：\n   - Robyn R 版依赖 Python 库 `nevergrad`。\n   - Robyn Python 原生版（Beta）需配置虚拟环境。\n\n> **注意**：国内用户若下载缓慢，可在 R 中设置清华或中科大镜像源：\n> ```r\n> options(repos = c(CRAN = \"https:\u002F\u002Fmirrors.tuna.tsinghua.edu.cn\u002FCRAN\"))\n> ```\n\n---\n\n## 安装步骤\n\n### 方案 A：安装 R 版本（推荐，功能最稳定）\n\n1. **安装 CRAN 正式版**：\n   ```r\n   install.packages(\"Robyn\")\n   ```\n\n2. **或安装 GitHub 开发版**（如需最新功能）：\n   ```r\n   # 若未安装 remotes 包，先运行：install.packages(\"remotes\")\n   remotes::install_github(\"facebookexperimental\u002FRobyn\u002FR\")\n   ```\n\n3. **常见问题处理**：\n   - **下载超时**：在 R 中运行 `options(timeout=400)` 后再安装。\n   - **Windows OpenSSL 错误**：需手动安装 OpenSSL 并配置环境变量（参考 StackOverflow 相关教程）。\n   - **Nevergrad 报错**：确保已安装 Python 并在系统中可用。\n\n### 方案 B：安装 Python 原生版本（Beta 测试版）\n\n> ⚠️ 此版本由 LLM 辅助翻译，可能存在 Bug，适合进阶开发者贡献代码或尝鲜。\n\n1. **创建并激活虚拟环境**：\n   - **Windows**:\n     ```bash\n     python -m venv robyn-env\n     robyn-env\\Scripts\\activate\n     ```\n   - **macOS\u002FLinux**:\n     ```bash\n     python3 -m venv robyn-env\n     source robyn-env\u002Fbin\u002Factivate\n     ```\n\n2. **安装 Robyn Python 包**：\n   ```bash\n   # 推荐从 PyPI 安装\n   pip3 install robynpy\n   \n   # 或从源码安装开发版\n   # pip3 install -r requirements.txt\n   ```\n\n### 方案 C：Python 调用 R 内核（API 封装版）\n\n如果不想直接使用 Python 重写版，可通过 `plumber` 接口在 Python 中调用稳定的 R 版本。需先按**方案 A**安装 R 包，具体用法参考官方 `robyn_api` 笔记。\n\n---\n\n## 基本使用\n\n### R 语言使用示例\n\n最快捷的方式是运行官方提供的演示脚本，该脚本包含模拟数据及完整流程（数据预处理、建模、聚类、预算分配等）。\n\n1. **加载包并运行演示**：\n   ```r\n   library(Robyn)\n   \n   # 获取演示脚本路径 (需先下载 demo.R 文件或在线查看)\n   # 建议直接复制官方 demo.R 内容到本地运行\n   ```\n\n2. **核心函数调用逻辑**（简化版）：\n   ```r\n   # 准备数据 (input_data 需符合特定格式，包含日期、响应变量、媒体花费等)\n   # 运行模型\n   robyn_model \u003C- robyn(\n     data = input_data,\n     date_var = \"ds\",\n     dep_var = \"y\",\n     media_vars = c(\"paid_search\", \"display\", \"video\"),\n     plot = TRUE,\n     output_folder = \"Robyn_Results\"\n   )\n   \n   # 查看结果\n   plot(robyn_model)\n   ```\n\n> **提示**：强烈建议下载并运行官方 [`demo.R`](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Ftree\u002Fmain\u002Fdemo\u002Fdemo.R) 脚本，它涵盖了绝大多数标准用例。\n\n### Python 原生版使用示例\n\n1. **运行官方教程 Notebook**：\n   进入安装目录下的 `python\u002Fsrc\u002Frobyn\u002Ftutorials` 文件夹。\n\n2. **一键运行（推荐新手）**：\n   打开 `tutorial1.ipynb`，该文件提供了端到端的流程：\n   - 特征工程\n   - 模型训练\n   - 聚类分析\n   - 生成单页报告 (One-pager)\n   - 预算分配建议\n   \n   直接在 Jupyter 中修改配置单元格并运行所有 Cell 即可。\n\n3. **模块化调用（进阶）**：\n   使用 `tutorial1_src.ipynb` 可单独调用特定模块（如跳过聚类），适合需要自定义工作流的用户。\n\n---\n\n**更多资源**：\n- 官方文档网站：[Robyn Site](https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002F)\n- 交流社区：[Facebook Group](https:\u002F\u002Fwww.facebook.com\u002Fgroups\u002Frobynmmm\u002F)\n- 官方课程：[Meta Blueprint](https:\u002F\u002Fwww.facebookblueprint.com\u002Fstudent\u002Fpath\u002F253121-marketing-mix-models)","某中型电商公司的数据分析师正面临季度营销复盘，需要评估社交媒体、搜索广告和邮件营销等十个渠道的真实贡献，以优化下一季度的百万级预算分配。\n\n### 没有 Robyn 时\n- **建模门槛极高**：传统营销组合模型（MMM）依赖昂贵的外部咨询或资深统计专家，内部团队因缺乏贝叶斯推断和时间序列分解的专业知识而无法独立开展。\n- **人工调参耗时且主观**：确定广告衰减率（Adstock）和饱和曲线需反复手动试错，不仅耗费数周时间，还极易引入人为偏见，导致结果不稳定。\n- **难以应对多维数据**：面对数字营销产生的细粒度日度数据和众多变量，传统方法计算效率低下，经常出现过拟合或无法收敛的情况。\n- **预算分配缺乏科学依据**：由于无法量化各渠道的边际效应，最终预算决策往往基于直觉或历史惯性，而非数据驱动的增量回报分析。\n\n### 使用 Robyn 后\n- **自动化降低技术壁垒**：Robyn 内置的岭回归和多目标进化算法自动完成超参数优化，让普通数据分析师也能在几天内构建出专业的 MMM 模型。\n- **消除人为偏差**：通过半自动化流程客观探索广告衰减和饱和形态，消除了主观猜测，确保模型结果可复现且更贴近真实市场反应。\n- **高效处理复杂数据**：专为拥有大量自变量的细粒度数据集设计，轻松处理数字广告的复杂波动，快速分解趋势与季节性因素。\n- **精准指导预算投放**：基于梯度优化算法直接输出各渠道的最佳预算分配方案，清晰展示每增加一元投入带来的预期转化增量。\n\nRobyn 将原本只有巨头玩得起的复杂建模能力 democratize（民主化），让中小型企业也能用科学数据驱动营销增长。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Ffacebookexperimental_Robyn_2f4dd8cb.png","facebookexperimental","Meta Experimental","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Ffacebookexperimental_449342bd.png","These are Meta projects that are not necessarily used in production but are being developed in the open nevertheless.",null,"https:\u002F\u002Fopensource.fb.com","https:\u002F\u002Fgithub.com\u002Ffacebookexperimental",[80,84,88,91,95,99,102],{"name":81,"color":82,"percentage":83},"Jupyter Notebook","#DA5B0B",92.6,{"name":85,"color":86,"percentage":87},"Python","#3572A5",3.8,{"name":89,"color":90,"percentage":10},"R","#198CE7",{"name":92,"color":93,"percentage":94},"MDX","#fcb32c",0.6,{"name":96,"color":97,"percentage":98},"JavaScript","#f1e05a",0,{"name":100,"color":101,"percentage":98},"CSS","#663399",{"name":103,"color":104,"percentage":98},"Dockerfile","#384d54",1445,425,"2026-04-18T04:51:09","MIT",4,"Linux, macOS, Windows","未说明",{"notes":113,"python":114,"dependencies":115},"1. 即使使用 Python 版本，机器上也必须预先安装 R 语言和 glmnet R 包。\n2. Windows 用户若遇到 openssl 错误，需手动安装并配置 OpenSSL。\n3. Python 版本目前为 Beta 版（由 LLM 翻译），可能存在 Bug。\n4. 另提供基于 R 包的 Python API 包装器方案作为替代。","未说明 (需安装 Python 以创建虚拟环境，具体版本未在 README 中明确指定)",[116,117,118,119],"R (必须安装)","glmnet (R 包)","nevergrad (Python 库)","robynpy (Python 包)",[14,121],"其他",[123,124,125,126,127,128,129,130,131,132,133,134],"marketing-mix-modeling","marketing-mix-modelling","mmm","marketing-science","econometrics","adstocking","cost-response-curve","budget-allocation","hyperparameter-optimization","evolutionary-algorithm","ridge-regression","gradient-based-optimisation","2026-03-27T02:49:30.150509","2026-04-19T03:05:02.888392",[138,143,148,153,158,163],{"id":139,"question_zh":140,"answer_zh":141,"source_url":142},41552,"运行 robyn_refresh 时遇到 'NMF::createStream - invalid value for n' 错误怎么办？","该错误通常发生在增量天数较多（如超过 22 天）或更改了目标变量后。维护者已确认并在后续版本中修复了此问题。建议将 Robyn 升级到最新版本（如 3.6.2 或更高），如果问题依旧，请尝试重新运行模型并选择与新版本兼容的候选模型生成新的 JSON 文件后再进行刷新。","https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fissues\u002F307",{"id":144,"question_zh":145,"answer_zh":146,"source_url":147},41553,"遇到 'subscript out of bounds' 错误或日期变量相关报错如何解决？","此错误常由媒体变量命名不规范（如包含多个下划线）或日期变量配置错误引起。首先检查媒体变量命名，确保只使用单个下划线（例如重命名为 '_s' 和 '_i'）。其次，检查 'date_var' 参数，确保只提供了一个正确的日期变量名。如果环境状态异常，尝试重启 R 会话并清除全局环境变量（clear global environment）后重新运行脚本。","https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fissues\u002F359",{"id":149,"question_zh":150,"answer_zh":151,"source_url":152},41554,"Robyn Refresh 功能失败或提示时间序列验证未激活如何处理？","这通常是由于用于刷新的 JSON 文件是由旧版本 Robyn 生成的，导致与新版本不兼容。解决方案是：使用最新版本的 Robyn 重新运行模型，选择一个与旧模型表现相当的候选模型，导出新的 JSON 文件，然后使用该新文件执行 recreate 和 refresh 操作。注意，由于包仍在开发中，向后兼容性并不总是保证。","https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fissues\u002F960",{"id":154,"question_zh":155,"answer_zh":156,"source_url":157},41555,"在 Budget Allocator 中使用 date_min 和 date_max 时出现 'missing value where TRUE\u002FFALSE needed' 错误？","当某些渠道的系数为 0 或在指定日期范围内数据缺失时，优化器可能会抛出此错误。如果遇到此问题，建议检查数据中是否存在零支出或非数值情况。维护者建议提供数据集和模型的 JSON 文件以便复现和调试。临时解决方法是尝试使用非零支出的占位数据测试该渠道，或调整日期范围以确保数据完整性。","https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fissues\u002F366",{"id":159,"question_zh":160,"answer_zh":161,"source_url":162},41556,"按照 Demo 教程安装 nevergrad 失败怎么办？","安装 nevergrad 失败通常与环境配置有关。确保已正确安装 reticulate 包，并尝试手动创建虚拟环境（virtualenv_create(\"r-reticulate\")）。如果通过 pip 安装失败，检查系统是否安装了 Python 及 pip 工具。对于 Windows 用户，可能需要以管理员身份运行 RStudio 或手动配置 Python 路径。如果问题持续，建议检查具体的 pip 报错日志以定位依赖缺失问题。","https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fissues\u002F189",{"id":164,"question_zh":165,"answer_zh":166,"source_url":147},41557,"Michaelis-Menten 拟合超出范围提示 'Using lm instead' 是什么意思？","这表示对于特定的媒体变量（如 fbprosp_I, yt_I 等），数据特征不符合 Michaelis-Menten 曲线的假设（例如饱和效应不明显），因此 Robyn 自动退化为使用线性模型（lm）进行拟合。这通常不是致命错误，但如果大量变量出现此提示，建议检查数据是否存在异常值、零值过多或变量命名是否正确（避免多余的下划线）。",[168,173,178,183,188,193,198,203,208,213,218,223,228,233,238,243,248,253,258,263],{"id":169,"version":170,"summary_zh":171,"released_at":172},333536,"v3.12.0","## 变更内容\n曝光拟合、曲线校准以及覆盖与频次分配器（https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F1132）\n* 功能：[启用曝光拟合](https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002Fdocs\u002Ffeatures#paid-media-variables)\n  - 在父模型拟合中，曝光量（Imp\u002FGRP 等）将优先于支出。\n  - 弃用 `fit_spend_exposure` 函数，包括 Michaelis-Menten 模型。支出与曝光之间的非线性拟合并未显著提升拟合效果。未来，曲线校准功能将专注于改进曲线识别。\n  - 使用 CPE（每次曝光成本）作为支出与曝光之间的转换比率，并通过 `cpe_window` 对整个数据集进行缩放，以获得适合建模周期的支出规模。\n  - 移除 `minpack.lm` \u002F `nlsLM` 依赖项。\n  - 更新曝光量图表。\n\n* 功能：[曲线校准器](https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002Fdocs\u002Ffeatures#holistic-calibration) - `robyn_calibrate`\n  - 使用频次桶模拟累积的 R&F 数据集。\n  - 在 Nevergrad 超参数中除了 `alpha` 和 `gamma` 外，新增 `beta` 系数，以提升曲线拟合效果。\n  - 绘制包含频次桶的图表，并为每次试验生成一页概览图。\n  - 添加 `df_curve_reach_freq` 作为示例数据集。\n  - 创建 `robyn_calibrate` 函数，接收曲线输入并输出超参数范围作为输入。\n  - 将之前的内部 `robyn_calibrate` 函数重命名为 `lift_calibration`。\n  - 使用 `while` 循环实现提前停止收敛。\n  - 更新文档。\n\n* 原型：[覆盖与频次分配器](https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002Fdocs\u002Ffeatures#the-reach--frequency-allocator-prototype)\n  这是一个包含以下内容的 R&F 分配器概念验证：\n  - 模拟的 R&F 数据。\n  - 使用 Nevergrad 的乘法方程估算 R&F Hill 参数。\n  - 曲面可视化。\n  - 基于 NLOPT 的 R&F 分配器。\n  - 约束验证。\n\n* 更新：检查、输入、转换及网站\n  - 简化各类检查函数。\n  - 调整 `model.R`，包括重置 `run_transformations` 参数，以便更清晰地了解所需参数。简化 `transformation.R`，移除不必要的检查。\n  - 在 `model.R` 和 `pareto.R` 中：从两个脚本中移除 `decompSpendDist`，以减少内存泄漏；改用 `xDecompAgg` 子集。\n  - 在 `transformation.R` 和 `response.R` 中：统一 `run_transformation` 和 `robyn_response` 中的转换命名。\n  - 在 `response.R` 中：移除曝光外推，因为已在 `robyn_input` 中完成。同时在输出中添加拐点信息。\n  - 在 `plots.R` 中：修复一页式饱和度图表的问题。\n  - 在 `pareto.R` 中：将 `run_dt_resp()` 重写为 `response_wrapper`，并对转换逻辑和命名进行统一。\n  - 在 `pareto.R` 中：用 `lapply` 替代 `foreach` 循环，以简化代码。\n  - 在 `pareto.R` 中：简化图表数据生成流程，尤其是饱和度曲线、实际值与预测值对比图以及即时效应与延滞效应对比图。\n  - 在 `pareto.R` 中：去除冗余的 `xDecompVecCollect`，移除 `rawMedia`、`rawSpend`、`predictedExposure`、`saturatedMedia` 和 `saturatedSpendReversed` 类型，仅保留用于响应计算的 `adstockedMedia` 和 `decompMedia`。","2024-12-19T13:49:33",{"id":174,"version":175,"summary_zh":176,"released_at":177},333537,"v3.11.1","## 变更内容\n* 修复：@laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F1011 中修复了单页反编译标签和警告。","2024-06-27T07:53:11",{"id":179,"version":180,"summary_zh":181,"released_at":182},333538,"v3.11.0","## 变更内容\n* 添加目标权重，由 @gufengzhou 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F826 中完成\n* build(deps): 将 \u002Fwebsite 中的 @babel\u002Ftraverse 从 7.21.5 升级到 7.23.2，由 @dependabot 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F842 中完成\n* 新特性：检查并修复 all_sol_json 和 new pareto_df 参数的输入，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F843 中完成\n* 新特性 + 修复：对 JSON 文件进行修复和改进，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F852 中完成\n* 修复：与 robyn_recreate() 中的旧版文件兼容性问题，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F865 中完成\n* 修复 #850 和 #838，由 @michellegrushkometa 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F864 中完成\n* Robyn API，由 @yu-ya-tanaka 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F863 中完成\n* 修复：当处理月度数据时，cut() 函数会崩溃的问题，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F868 中完成\n* 修复：当某一列的所有值均为负数时，程序会崩溃的问题，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F871 中完成\n* 修复：在 na.rm = TRUE 的情况下，winsorize 函数出现错误的问题 #872；以及在未使用 Prophet 时，InputCollect 打印信息的问题，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F875 中完成\n* 一系列小的修复和新功能，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F882 中完成\n* 修复：为提高可重复性，添加 foreach 种子；同时新增 dt_simulated_weekly 数据集，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F907 中完成\n* 更新 Robyn API 笔记本和端点，由 @yu-ya-tanaka 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F912 中完成\n* 修复：避免在场景变更时覆盖 CSV 文件，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F913 中完成\n* 修复：在输出中加入截距参数，并添加 ts_validation 图以验证收敛性，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F917 中完成\n* build(deps): 将 \u002Fwebsite 中的 follow-redirects 从 1.15.3 升级到 1.15.4，由 @dependabot 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F899 中完成\n* 新特性：为 decomp_plot() 启用 baseline_level 参数；当仅有一个模型可用时，允许 solID 为 NULL；更新 robyn_response() 的示例文档，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F923 中完成\n* 文档：澄清 objective_weights 的顺序，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F929 中完成\n* 新特性：在单页报告中添加聚类信息，修复指标说明，减小文本尺寸，移除 ng 检查以支持重新创建模型；同时修复 RNG 警告，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F928 中完成\n* 新特性：在重新创建模型时，保留从 JSON 导出的自定义数据；如果原始数据可用，则优先使用原始数据；同时更新文档，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F930 中完成\n* 修复：mediaVecCollect 的有机变量未乘以系数值的问题；以及在未设置 Prophet 国家时，InputCollect 打印信息的问题，由 @laresbernardo 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F936 中完成\n*","2024-06-20T11:09:21",{"id":184,"version":185,"summary_zh":186,"released_at":187},333539,"v3.10.5","* **新功能**: 在 `robyn_run()` 中新增参数 `objective_weights`，用于手动调整目标函数（NRMSE、DECOMP.RSSD、MAPE.LIFT）的权重。默认权重为均等权重 c(1,1,1)。注意：此功能尚处于实验阶段，目前尚未提供关于如何通过权重来影响模型效果的指导。提交记录 [在此](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcommit\u002F7e581b775b37935c17b8d54d83b40a829429c873)\n* **新功能**: 元MMM API 连接器演示，一个测试版脚本。提交记录 [在此](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcommit\u002Fa6af850efd325298593941f3abe395fcf1b2b9d3)\n* **文档**: 更新并重构了官网的“功能”页面，同时重新组织了导航结构。详情请见 [此处](https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002Fdocs\u002Ffeatures)\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.10.3...v3.10.5","2023-10-24T03:00:32",{"id":189,"version":190,"summary_zh":191,"released_at":192},333540,"v3.10.3","关于 Facebook 的 Robyn 社区的更多详情 [帖子](https:\u002F\u002Fwww.facebook.com\u002Fgroups\u002Frobynmmm\u002Fposts\u002F1419605262140936\u002F)\n\n* **功能**：对象大小最多可减少 88% #687\n* **功能**：对无媒体效应的媒体（参数）施加 RSSD 惩罚 #680\n* **功能**：始终将预测系数约束在 0 到 1 之间 #686\n* **功能**：在多目标优化图表上使用温莎化 NRMSE [默认阈值为 1\u002F500]，以避免极端偏斜 #693\n* **功能**：新增 `intercept` 参数传递给 `glmnet()` #722\n* **修复**：当按建模窗口筛选后，变量无方差时，返回有意义的错误信息 #619\n* **修复**：移除了 Weibull PDF adstock 的负累积输出 #706\n* **修复**：在 `robyn_refresh()` 中正确传递 `plot_folder` 输入 #708\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.10.2...v3.10.3","2023-05-17T17:19:54",{"id":194,"version":195,"summary_zh":196,"released_at":197},333541,"v3.10.2","- **修复**：运行 `robyn_refresh()` 时，应根据 error_score 值正确选择刷新模型 #674\n- **修复**：在删除截距项时，也传递 `penalty.factor` 参数\n- **修复**：导入节假日数据时，强制将日期值转换为日期类型 #663，由 @richin13 提供\n- **文档**：更新了多个函数及网站的相关文档\n\n## 新贡献者\n* @richin13 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F663 中完成了首次贡献\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.10.1...v3.10.2","2023-04-14T14:03:43",{"id":199,"version":200,"summary_zh":201,"released_at":202},333542,"v3.10.1","更多详情请参阅文章：[使用 Robyn 预算分配器达成 ROAS 目标](https:\u002F\u002Fmedium.com\u002F@gufengzhou\u002Fhitting-roas-target-using-robyns-budget-allocator-274ace3add4f)\n\n* **功能**：分配器新增针对 ROAS\u002FCPA 目标的场景 “target_efficiency” #648\n* **修复**：修复分配器可视化表格中 CPA 值倒置的问题 #640\n* **修复**：重新启用实验性功能 add_penalty_factor\n* **修复**：重新计入被跳过的渠道的花费 #645\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.10.0...v3.10.1","2023-03-27T14:21:45",{"id":204,"version":205,"summary_zh":206,"released_at":207},333543,"v3.10.0","更多详情请参见文章：[Robyn 预算分配中边际 ROAS 的收敛性](https:\u002F\u002Fmedium.com\u002F@gufengzhou\u002Fthe-convergence-of-marginal-roas-in-the-budget-allocation-in-robyn-5d407aebf021)\n\n* **功能**：为 `robyn_allocator()` 新增了一张全新的单页概览，展示了初始、有预算上限以及无预算上限三种场景，默认使用上个月的数据。与之前版本相比的主要改动包括：初始支出现为所选日期范围内的平均值，而非非零平均值；废弃了“max_response_expected_spend”场景；曲线图中新增了结转信息；当预算已用完且无法完全分配时会提示用户；增加了 mROAS\u002FmCPA 指标，以便更好地理解分配情况。此外，这也使预测功能又向前迈进了一步。#600\n* **功能**：`robyn_response()` 现在需要指定日期或日期范围来进行广告库存衰减计算（默认为上一周期），并支持传入单个或多个值，以返回不同的使用场景和结果。\n* **功能**：新增了导出的包装函数 `transform_adstock()`。\n* **功能**：在测试集上增加了 NRMSE 验证。\n* **功能**：添加了 Prophet 的月度成分。\n* **修复**：解决了 `OutputCollect$OutputModels` 和 `OutputModels` 之间差异导致无法生成 `ts_validation` 图的问题。#596\n* **修复**：为固定超参数正确设置了 solID，不再使用 1_1_1。\n* **重构**：将 `OutputCollect` 中的 `xDecompVec` 大小缩减为仅包含帕累托前沿模型。\n* **重构**：移除了对 “ggcorrplot” 和 “rPref” 包的依赖。\n* **文档**：在 demo.R 中添加了蓝图链接。\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.9.0...v3.10.0","2023-02-28T15:29:58",{"id":209,"version":210,"summary_zh":211,"released_at":212},333544,"v3.9.0","- **功能**：新增时间序列验证功能，通过时间序列的动态训练\u002F验证\u002F测试集划分，并为每个分组特征报告调整后的 R² 和 NRMSE 指标 #545。我们添加了一个额外的 `train_size` 超参数，用于指定训练集的大小，默认会在 0.5–0.8 的范围内迭代。由于这是一个超参数，您可以手动更改范围或固定其值。使用 `robyn_run()` 中的新参数 `ts_validation` 来开启或关闭此功能；目前默认设置为 `FALSE`。这是即将推出的预测功能的重要一步。\n- **功能**：新增 `ts_validation()` 函数，用于绘制时间序列验证和收敛结果。当 `ts_validation = TRUE` 且 `export = TRUE` 时，会自动生成并导出名为 `ts_validation_plot.png` 的文件。\n- **修复**：更新了时间序列验证中调整后的 R² 计算方法 (`get_rsq()`)，使其使用相同的分母。\n- **修复**：现在不再按误差最小排序结果，以保持迭代结果的原始顺序。\n- **功能**：添加了 Prophet 的月度成分，以丰富分解结果 #525。\n- **修复**：为重新创建的固定超参数模型修正了 solID（不再是“1_1_1”）。\n- **代码重构**：在 `OutputCollect` 中仅保留帕累托前沿模型，从而减小了 `xDecompVec` 的规模。\n- **文档**：修改了 [demo.R](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fblob\u002Fmain\u002Fdemo\u002Fdemo.R) 文件中的标准输入，将建模窗口的数据范围扩大至更多数据（默认为 3 年）。\n\n## 新贡献者\n* @michellegrushkometa 在 https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F559 中完成了首次贡献。\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.8.2...v3.9.0","2022-12-15T15:16:43",{"id":214,"version":215,"summary_zh":216,"released_at":217},333545,"v3.8.2","- **新功能**：为每个试验的帕累托前沿模型新增状态栏，用于显示计算状态信息\n- **新功能**：将累积结果纳入 `pareto_aggregated.csv` 输出文件以及 `OutputCollect$xDecompAgg$carryover_pct` 中\n- **新功能**：新增错误提示，明确指出缺少哪些超参数输入 #543\n- **修复**：通过移除输出中的冗余和未使用数据，大幅缩减了 `robyn_run()` 和 `robyn_outputs()` 结果的大小（相比 3.8.1 版本约减少 80%）#534\n- **修复**：修复了 `check_factorvars()` 中的无效参数类型问题，以及重新创建校准模型时出现的问题 #520\n- **修复**：`add_penalty_factor` 参数现可正确与 JSON 文件及 `robyn_refresh()` 配合使用 #543\n- **修复**：修正自定义数据的超参数长度问题 #533\n- **修复**：修复了 RobynLearn 在检查数值型数据时存在的 bug #532\n- **修复**：移除了旧版演示 `.RData` 文件中的 `.iData` 格式\n- **修复**：现在传入自定义 `pareto_fronts` 输入而非使用“auto”选项，行为符合预期\n- **文档**：更新了官网、meta.com 邮件中的发布版本信息，并在 `robyn_update()` 中更新了 CRAN 链接\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.8.0...v3.8.2","2022-11-23T17:30:36",{"id":219,"version":220,"summary_zh":221,"released_at":222},333546,"v3.8.0","- **Feat:** Added in-cluster bootstrapped confidence intervals (CI) for ROAS and CPA. We treat each cluster of Pareto-optimal model candidates as a sample from a local optimum of the entire population. Default parameters can be customized manually with `boot_n` and `sim_n` arguments.\r\n- **Feat:** New `robyn_calibrate()` function that replaces  previous un-exported function `calibrate_mmm()`. The new calibration method is able to separate immediate & carryover effects. When calibrating using experimental results, only the immediate response and its future carryover serve as a calibration target, as opposed to previously the total response. The historical response is excluded from calibration.\r\n- **Feat**: Enabled multi-channel calibration so we can use experiments that measured more than one channel with a single experiment to be used for calibration (i.e. incrementality experiment measured all `fb` but you had `fb_brand` and `fb_perf` as two separate media channels\u002Fvariables).\r\n- **Feat:** Added 2 new plots into model one-pager: bootstrapped CI plot and immediate vs carryover response plot.\r\n- **Feat**: Changed default Pareto-fronts from `3`  to `”auto\"` to pick the N that contains at least 100 models (threshold can be changed manually with `min_candidates` parameter).\r\n- **Recode**: improved CodeFactor's code quality score from C- to A\r\n- **Feat**: Additional CI outputs containing revamped plot and CSV file.\r\n- **Feat**: Enabled turning off parallel calculations when `cores = 1`.\r\n- **Fix**: Fixed few minor bugs and doumentations (#496, #506, #507, #515)\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.7.2...v3.8.0","2022-10-28T07:50:36",{"id":224,"version":225,"summary_zh":226,"released_at":227},333547,"v3.7.2","- **Feat**: wrap `robyn_mmm()` with a `tryCatch()` to return partial results if the function crashes after a certain time running and warns the user when this happens\r\n- **Feat**: auto-detect categorical variables (no need to set `factor_vars` parameter in `robyn_inputs()`)\r\n- **Feat**: include R and Robyn's versions to JSON files and InputCollect for reproducibility\r\n- **Feat**: export\u002Fsave raw data input for reproducibility (raw_data.csv file)\r\n- **Feat**: set `Robyn::dt_prophet_holidays` as default input on `dt_holidays` parameters\r\n- **Fix**: inverted counters in `check_hyperparameters()` message #474\r\n- **Fix**: force date format before binding rows in `robyn_refresh()` #480\r\n- **Fix**: `check_context()` was being skipped in some cases\r\n- **Fix**:  when only 1 categorical value with 2 unique values crashed one-hot-encoding\r\n- **Docs**: updated templates for issues and pull requests\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.7.1...v3.7.2\r\n**Full Changelog since last CRAN update**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.6.3...v3.7.2","2022-09-01T20:15:51",{"id":229,"version":230,"summary_zh":231,"released_at":232},333548,"v3.7.1","- **Feat**: new `robyn_read()` and `robyn_write()` functions to save and load Robyn models in a transparent, flexible, and cost-efficient way using JSON instead of RDS files (read [more](https:\u002F\u002Fwww.facebook.com\u002Fgroups\u002Frobynmmm\u002Fposts\u002F1252761488825315\u002F)); also, new `print` methods for both objects containing the most relevant information\r\n-  **Feat**: new `robyn_recreate()` to rebuild any model's `InputCollect` and `OutputCollect` objects based on their JSON files and data\r\n- **Feat**: reactivated spend exposure fitting and plotting #463\r\n- **Feat**: updated `robyn_response()` to receive numeric vectors #464\r\n- **Feat**: enabled `calibration_input` on `robyn_refresh()` to calibrate on the fly and more robust checks on data inputs\r\n- **Feat** added Robyn and R versions as the caption in one-pagers to help users debug\r\n- **Feat**: trimmed spend response curves on `robyn_allocator()` and `robyn_onepagers()` plots outputs\r\n- **Fix**: missed intercept calculation in fitted vs residual plot #462\r\n- **Fix**: when single categorical value had 2 levels it crashed the one-hot-encoding process\r\n- **Fix**: datasets with no categorical data crashed when using one-hot-encoding #419\r\n- **Fix**: no need to manually sort the dates before passing the data to `robyn_inputs()`. Ref: `check_datevar()` #448\r\n- **Fix**: fixed ggplot warnings on some plots (previously hidden with suppressWarnings)\r\n- **Other**: added badges with website and Facebook group in README files (see [here](https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn#readme)), updated documentation and website, and more data checks on user inputs\r\n\r\n## New Contributors\r\n* @Tomobay made his first contribution #464\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.7.0...v3.7.1","2022-08-26T15:38:12",{"id":234,"version":235,"summary_zh":236,"released_at":237},333549,"v3.7.0","## Relevant changes on v3.7.0:\r\n\r\n- **Recode**: got rid of data.table dependency for r2py wrapper and removed all `globalVariables` associated noise\r\n- **Recode**: all code is now clean and formatted under the tidyverse code style for better code reading and standardization\r\n- **Feat**: trimmed functionality for response curves on one-pagers outputs to have coherent ranges plotted\r\n- **Feat**: enabled channels removal on `robyn_allocator()` by setting their constraints to 0 #411\r\n- **Feat**: when manually selecting refresh models in `robyn_refresh()`, re-ask user until valid solID is provided, instead of crashing\r\n- **Feat**: new `plot` and improved `print` methods for `robyn_refresh()` outputs\r\n- **Feat**: include time units used in adstock plots for clarity\r\n- **Feat**: enabled organic media variables to be calibrated (no spend)\r\n- **Fix**: when best model based on minimum combined errors was tied with other models, inconsistent outputs (one-pagers \u002F clustering). Standardized combined errors methodology with new `errors_scores()` function, especially normalizing errors before filtering models. The largest the \"error_score\", the better the model's performance #428\r\n- **Fix**: show blue dots on top of grey dots in Pareto plots #420\r\n- **Fix**:  positive\u002Fnegative colour palette on waterfall plot when all values are positive\r\n- **Fix**:  set prophet's print as disabled when prophet_vars input is NULL (off)\r\n- **Docs**: added CRAN, site, and FB group badges on README files\r\n- **Docs**: several typos and documentation updates\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.6.3...v3.7.0","2022-07-27T19:18:23",{"id":239,"version":240,"summary_zh":241,"released_at":242},333550,"v3.6.3","## Relevant changes on v3.6.3:\r\n* **CRAN**: First Robyn version available via CRAN. From now on, install CRAN's for stable version, GitHub's for dev version.\r\n* **Docs**: Site revamp #372, documentation updates and demo enrichment\r\n* **Feat**: Added `version_prompt` parameter to robyn_refresh() #375\r\n* **Feat**: Added new calibration checks to ensure quality experiments usage\r\n* **Feat**: New `date_min` and `date_max` parameters on `robyn_allocator()` to pick non-0 means window\r\n* **Feat**: New `robyn_update()` function\r\n* **Feat**: More checks and warnings included to push users to follow best practices.\r\n* **Refactor**: Changed 1 to 3 Pareto fronts as default to enrich `robyn_clusters()` results\r\n* **Refactor**: Changed default thresholds on `robyn_converge()` to be more flexible\r\n* **Fix**: Several bugs squashed \r\n\r\n## New Contributors\r\n* @mast4461 made their first contribution in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F362\r\n* @JustStas made their first contribution in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F375\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.6.2...v3.6.3","2022-05-06T14:57:13",{"id":244,"version":245,"summary_zh":246,"released_at":247},333551,"v3.6.2","## Relevant changes on v3.6.2:\r\n- **Viz**: removed redundant information on plots and standardized styles and contents on all visualizations.\r\n- **Feat**: new `date_min` and `date_max` parameters on `robyn_allocator()` to pick specific date range to consider mean spend values ([user request](https:\u002F\u002Fwww.facebook.com\u002Fgroups\u002Frobynmmm\u002Fpermalink\u002F1072870463481086)).\r\n- **Feat**: new `plot` methods for `robyn_allocator()` and `robyn_save()` outputs, and `print` method for `robyn_inputs()` with and without hyperparameters.\r\n- **Feat**: provide recommendations on calibration inputs depending on the experiments' confidence, spending, and KPI measured (#307).\r\n- **Feat**: warn and avoid weekly trend input when data granularity is larger than \"week\".\r\n- **Fix**: issues on several `robyn_allocator()` specific cases (#349, #344, #345), especially when some coefficients were 0.\r\n- **Fix**: bug with Weibull adstock scenario (#353).\r\n- **Docs**: fixed some typos, updated, and standardized internal documentation. \r\n\r\n## Commits log\r\n* Fix jde 20211209 by @jeffedwards in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F234\r\n* Added a couple file to run the code with actual data by @jeffedwards in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F235\r\n* Jde 20211209 general and troubleshoot by @jeffedwards in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F236\r\n* quick updates by @jeffedwards in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F237\r\n* quick updates by @jeffedwards in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F238\r\n* Updated for fix object to work. Updating check conditions to work by @jeffedwards in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F254\r\n* build(deps): bump follow-redirects from 1.14.7 to 1.14.8 in \u002Fwebsite by @dependabot in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F302\r\n* Fix by @jeffedwards in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F303\r\n* Revert \"Fix\" by @jeffedwards in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F304\r\n* Updated with master branch.  Created new Robyn file by @jeffedwards in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F305\r\n* PR v3.6.0  by @gufengzhou in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F314\r\n* build(deps): bump url-parse from 1.5.3 to 1.5.7 in \u002Fwebsite by @dependabot in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F311\r\n* build(deps): bump prismjs from 1.25.0 to 1.27.0 in \u002Fwebsite by @dependabot in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F326\r\n* build(deps): bump url-parse from 1.5.7 to 1.5.10 in \u002Fwebsite by @dependabot in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F327\r\n* Weibull testfix by @kyletgoldberg in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F355\r\n* Fix log message \"Using custom prophet parameters\" by @andrey-legayev in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F356\r\n* build(deps): bump minimist from 1.2.5 to 1.2.6 in \u002Fwebsite by @dependabot in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F357\r\n* Allocation and plots improvements. Version 3.6.2 by @laresbernardo in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F361\r\n\r\n## New Contributors\r\n* @kyletgoldberg made their first contribution in https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fpull\u002F355\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.6.0...v3.6.2","2022-03-31T16:46:26",{"id":249,"version":250,"summary_zh":251,"released_at":252},333552,"v3.6.0","### What's New in v3.6.0\r\n\r\n- New **hyperparameter \"lambda\"** finds MOO-optimal lambda and thus removes the need of manual lambda selection.\r\n- New optional **hyperparameter `penalty.factor`** that further extends hyperparameter spaces and thus potentially better fit.\r\n- New **optimisation convergence rules & plots** for each objective function showing if set iterations have converged or not (NRMSE, DECOMP.RSSD, and MAPE if calibrated)\r\n- Improved **response function** now also returns the response for exposure metrics (response on imps, GRP, newsletter sendings, etc) and plots. Note that argument names and output class has changed. See updated `demo.R` for more details.\r\n- More **budget allocation stability** by defaulting fitting media variables from `paid_media_vars` to `paid_media_spends`. Spend exposure fitting with Michaelis Menten function will only serve `robyn_response()` function output and plotting. `robyn_allocator()` now only relies on direct spend - response transformation.\r\n- Default **beta coefficient signs**: positive for paid & organic media and unconstrained for the rest. Users can still set signs manually.\r\n- New **print methods** for `robyn_inputs()`, `robyn_run()`, `robyn_outputs()`, and `robyn_allocator()` outputs to enable visibility on each step's results and objects content.\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.5.1...v3.6.0","2022-02-22T11:42:40",{"id":254,"version":255,"summary_zh":256,"released_at":257},333553,"v3.5.0","## What's New in v3.5.0\r\n- New `robyn_clusters()` function to reduce the number of models to select from after Pareto front solutions are picked\r\n- Auto-select K clusters given a minimum WSS variance on `robyn_clusters()`\r\n- Split `robyn_run()` functionalities into `robyn_outputs()` for more control over modelings results and exporting process\r\n- Enabled custom prophet inputs to be used in modeling and refreshing models\r\n- Enabled users to use all outputs without the need of exporting results (Shiny devs)\r\n- New `hyper_limits()` helper function with permitted hyper-parameters limits\r\n- New quiet mode to reduce prints and messages, few recoding, and overall improvements\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Ffacebookexperimental\u002FRobyn\u002Fcompare\u002Fv3.4.8...v3.5.0","2022-02-01T16:04:25",{"id":259,"version":260,"summary_zh":261,"released_at":262},333554,"v3.4.8","- 20% - 40% faster in the major modelling loop\r\n- 50% fasterr in plotting loop for linux & windows users","2021-11-25T19:48:21",{"id":264,"version":265,"summary_zh":76,"released_at":266},333555,"v3.4.4","2021-11-03T13:46:26"]