[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-GPflow--GPflow":3,"tool-GPflow--GPflow":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},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。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},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",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},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。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":67,"owner_name":67,"owner_avatar_url":75,"owner_bio":76,"owner_company":77,"owner_location":77,"owner_email":77,"owner_twitter":77,"owner_website":78,"owner_url":79,"languages":80,"stars":89,"forks":90,"last_commit_at":91,"license":92,"difficulty_score":23,"env_os":93,"env_gpu":94,"env_ram":93,"env_deps":95,"category_tags":101,"github_topics":102,"view_count":10,"oss_zip_url":77,"oss_zip_packed_at":77,"status":16,"created_at":114,"updated_at":115,"faqs":116,"releases":146},779,"GPflow\u002FGPflow","GPflow","Gaussian processes in TensorFlow","GPflow 是一个基于 Python 的开源库，专门用于构建高效的高斯过程模型。对于机器学习研究者来说，高斯过程是处理不确定性问题的利器，但传统实现往往计算缓慢且代码繁琐。GPflow 正是为了解决这些痛点而生。\n\n它底层依托 TensorFlow 2.4+ 和 TensorFlow Probability，充分利用 GPU 加速计算，显著提升了训练速度。其设计亮点在于支持灵活组合的核函数与似然函数，让模型搭建更加模块化。无论是贝叶斯优化还是概率回归，GPflow 都能提供现代化的推断框架。\n\nGPflow 非常适合机器学习工程师、数据科学家以及从事概率建模的研究人员使用。社区活跃且文档详尽，无论你是想快速上手实验，还是希望参与开源贡献，都能在这里找到支持。如果你正在寻找一个稳定、高性能的概率编程工具，GPflow 值得尝试。","\u003Cdiv style=\"text-align:center\">\n\u003Cimg width=\"500\" height=\"200\" src=\"https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fblob\u002Fdevelop\u002Fdoc\u002Fsphinx\u002F_static\u002Fgpflow_logo.svg\">\n\u003C\u002Fdiv>\n\n[![CircleCI](https:\u002F\u002Fcircleci.com\u002Fgh\u002FGPflow\u002FGPflow\u002Ftree\u002Fdevelop.svg?style=svg)](https:\u002F\u002Fcircleci.com\u002Fgh\u002FGPflow\u002FGPflow\u002Ftree\u002Fdevelop)\n[![Coverage Status](http:\u002F\u002Fcodecov.io\u002Fgithub\u002FGPflow\u002FGPflow\u002Fcoverage.svg?branch=master)](http:\u002F\u002Fcodecov.io\u002Fgithub\u002FGPflow\u002FGPflow?branch=master)\n[![Slack Status](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fslack-gpflow-green.svg?logo=Slack)](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fgpflow\u002Fshared_invite\u002FenQtOTE5MDA0Nzg5NjA2LTYwZWI3MzhjYjNlZWI1MWExYzZjMGNhOWIwZWMzMGY0YjVkYzAyYjQ4NjgzNDUyZTgyNzcwYjAyY2QzMWRmYjE)\n\n\n[Website](https:\u002F\u002Fgpflow.org) |\n[Documentation (release)](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002F) |\n[Documentation (develop)](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop) |\n[Glossary](GLOSSARY.md)\n\n#### Table of Contents\n\u003C!-- created with help from https:\u002F\u002Fgithub.com\u002Fekalinin\u002Fgithub-markdown-toc and further manual adjustments -->\n\n* [What does GPflow do?](#what-does-gpflow-do)\n* [Installation](#installation)\n* [Getting Started with GPflow 2.0](#getting-started-with-gpflow-20)\n* [The GPflow Community](#the-gpflow-community)\n   * [Getting help](#getting-help)\n   * [Slack workspace](#slack-workspace)\n   * [Contributing](#contributing)\n   * [Projects using GPflow](#projects-using-gpflow)\n* [Version Compatibility](#version-compatibility)\n   * [TensorFlow 1.x and GPflow 1.x](#tensorflow-1x-and-gpflow-1x)\n* [Citing GPflow](#citing-gpflow)\n\n\n## What does GPflow do?\n\nGPflow is a package for building Gaussian process models in Python.\nIt implements modern Gaussian process inference for composable kernels and likelihoods.\n\nGPflow builds on [TensorFlow 2.4+](http:\u002F\u002Fwww.tensorflow.org) and [TensorFlow Probability](https:\u002F\u002Fwww.tensorflow.org\u002Fprobability\u002F) for running computations, which allows fast execution on GPUs.\n\nThe [online documentation (latest release)](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002F)\u002F[(develop)](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop) contains more details.\n\n\n### Maintainers\n\nIt was originally created by [James Hensman](http:\u002F\u002Fjameshensman.github.io\u002F) and [Alexander G. de G. Matthews](https:\u002F\u002Fgithub.com\u002Falexggmatthews).\nIt is now actively maintained by (in alphabetical order)\n[Artem Artemev](http:\u002F\u002Fgithub.com\u002Fawav\u002F),\n[Mark van der Wilk](https:\u002F\u002Fmarkvdw.github.io\u002F),\n[ST John](https:\u002F\u002Fgithub.com\u002Fst--),\nand [Vincent Dutordoir](https:\u002F\u002Fvdutor.github.io\u002F).\nGPflow would not be the same without the community. **We are grateful to [all contributors](CONTRIBUTORS.md) who have helped shape GPflow.**\n\n *GPflow is an open source project. If you have relevant skills and are interested in contributing then please do contact us (see [\"The GPflow community\" section](#the-gpflow-community) below).*\n\n\n## Installation\n\n### Requirements\n\nGPflow depends on both TensorFlow (TF, version ≥ 2.4) and TensorFlow Probability (TFP, version ≥ 0.12). We support Python ≥ 3.7.\n\n**NOTE:** TensorFlow Probability releases are tightly coupled to TensorFlow, e.g. TFP 0.14 requires TF>=2.6, TFP 0.13 requires TF>=2.5, and TFP 0.12 requires TF>=2.4. Unfortunately, this is _not_ specified in TFP's dependencies. So if you already have an (older) version of TensorFlow installed, GPflow will pull in the latest TFP, which will be incompatible. If you get errors such as `ImportError: This version of TensorFlow Probability requires TensorFlow version >= 2.4`, you have to either upgrade TensorFlow (`pip install -U tensorflow`) or manually install an older version of the `tensorflow_probability` package.\n\n### Latest (stable) release from PyPI\n\n```bash\npip install gpflow\n```\n\n### Latest (bleeding-edge) source from GitHub\n\n*Be aware that the `develop` branch may change regularly, and new commits may break your code.*\n\nIn a check-out of the `develop` branch of the [GPflow GitHub repository](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow), run\n```bash\npip install -e .\n```\n\nAlternatively, you can install the latest GitHub `develop` version using `pip`:\n```bash\npip install git+https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow.git@develop#egg=gpflow\n```\nThis will automatically install all required dependencies.\n\n## Getting Started with GPflow 2.0\n\nThere is an [\"Intro to GPflow 2.0\"](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop\u002Fnotebooks\u002Fintro_to_gpflow2.html) Jupyter notebook; check it out for details.\nTo convert your code from GPflow 1 check the [GPflow 2 upgrade guide](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop\u002Fnotebooks\u002Fgpflow2_upgrade_guide.html).\n\n\n## The GPflow Community\n\n### Getting help\n\n**Bugs, feature requests, pain points, annoying design quirks, etc:**\nPlease use [GitHub issues](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fissues\u002F) to flag up bugs\u002Fissues\u002Fpain points, suggest new features, and discuss anything else related to the use of GPflow that in some sense involves changing the GPflow code itself.\nYou can make use of the [labels](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels) such as [`bug`](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels\u002Fbug), [`discussion`](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels\u002Fdiscussion), [`feature`](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels\u002Ffeature), [`feedback`](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels\u002Ffeedback), etc.\nWe positively welcome comments or concerns about usability, and suggestions for changes at any level of design.\n\nWe aim to respond to issues promptly, but if you believe we may have forgotten about an issue, please feel free to add another comment to remind us.\n\n**\"How-to-use\" questions:**\nPlease use [Stack Overflow (gpflow tag)](https:\u002F\u002Fstackoverflow.com\u002Ftags\u002Fgpflow) to ask questions that relate to \"how to use GPflow\", i.e. questions of understanding rather than issues that require changing GPflow code. (If you are unsure where to ask, you are always welcome to open a GitHub issue; we may then ask you to move your question to Stack Overflow.)\n\n### Slack workspace\n\nWe have a public [GPflow slack workspace](https:\u002F\u002Fgpflow.slack.com\u002F). Please use this [invite link](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fgpflow\u002Fshared_invite\u002FenQtOTE5MDA0Nzg5NjA2LTYwZWI3MzhjYjNlZWI1MWExYzZjMGNhOWIwZWMzMGY0YjVkYzAyYjQ4NjgzNDUyZTgyNzcwYjAyY2QzMWRmYjE) if you'd like to join, whether to ask short informal questions or to be involved in the discussion and future development of GPflow.\n\n### Contributing\n\nAll constructive input is gratefully received. For more information, see the [notes for contributors](CONTRIBUTING.md).\n\n### Projects using GPflow\n\nProjects building on GPflow and demonstrating its usage are listed below. The following projects are based on the current GPflow 2.x release:\n\n| Project | Description |\n| --- | --- |\n| [Trieste](https:\u002F\u002Fgithub.com\u002Fsecondmind-labs\u002Ftrieste)       | Bayesian optimization with TensorFlow, with out-of-the-box support for GPflow (2.x) models. |\n| [VFF](https:\u002F\u002Fgithub.com\u002Fst--\u002FVFF)       | Variational Fourier Features for Gaussian Processes (GPflow 2.x version) |\n| [BranchedGP](https:\u002F\u002Fgithub.com\u002FManchesterBioinference\u002FBranchedGP) | Gaussian processes with branching kernels.|\n| [VBPP](https:\u002F\u002Fgithub.com\u002Fst--\u002Fvbpp) | Implementation of \"Variational Bayes for Point Processes\".|\n| [Gaussian Process Regression on Molecules](https:\u002F\u002Fmedium.com\u002F@ryangriff123\u002Fgaussian-process-regression-on-molecules-in-gpflow-ee6fedab2130) | GPs to predict molecular properties by creating a custom-defined Tanimoto kernel to operate on Morgan fingerprints |\n\nIf you would like your project listed here, let us know - or simply [open a pull request](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fcompare) that adds your project to the table above!\n\n*The following projects build on older versions of GPflow (pre-2020); we encourage their authors to upgrade to GPflow 2.*\n\n| Project | Description |\n| --- | --- |\n| [GPflowOpt](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflowOpt)       | Bayesian Optimization using GPflow (stable release requires GPflow 0.5). |\n| [Doubly-Stochastic-DGP](https:\u002F\u002Fgithub.com\u002FICL-SML\u002FDoubly-Stochastic-DGP)| Deep Gaussian Processes with Doubly Stochastic Variational Inference.|\n| [widedeepnetworks](https:\u002F\u002Fgithub.com\u002Fwidedeepnetworks\u002Fwidedeepnetworks) | Measuring the relationship between random wide deep neural networks and GPs.|\n| [orth_decoupled_var_gps](https:\u002F\u002Fgithub.com\u002Fhughsalimbeni\u002Forth_decoupled_var_gps) | Variationally sparse GPs with orthogonally decoupled bases|\n| [kernel_learning](https:\u002F\u002Fgithub.com\u002Ffrgsimpson\u002Fkernel_learning) | Implementation of \"Differentiable Compositional Kernel Learning for Gaussian Processes\".|\n| [DGPs_with_IWVI](https:\u002F\u002Fgithub.com\u002Fhughsalimbeni\u002FDGPs_with_IWVI) | Deep Gaussian Processes with Importance-Weighted Variational Inference|\n| [kerndisc](https:\u002F\u002Fgithub.com\u002FBracketJohn\u002FkernDisc) | Library for automated kernel structure discovery in univariate data|\n| [Signature covariances](https:\u002F\u002Fgithub.com\u002Ftgcsaba\u002FGPSig) | kernels for (time)series as *inputs* |\n| [Structured-DGP](https:\u002F\u002Fgithub.com\u002Fboschresearch\u002FStructured_DGP) | Adding more structure to the variational posterior of the Doubly Stochastic Deep Gaussian Process |\n\n## Version Compatibility\n\nGPflow heavily depends on TensorFlow and as far as TensorFlow supports forward compatibility, GPflow should as well. The version of GPflow can give you a hint about backward compatibility. If the major version has changed then you need to check the release notes to find out how the API has been changed.\n\nUnfortunately, there is no such thing as backward compatibility for GPflow _models_, which means that a model implementation can change without changing interfaces. In other words, the TensorFlow graph can be different for the same models from different versions of GPflow.\n\n### TensorFlow 1.x and GPflow 1.x\n\nWe have stopped development and support for GPflow based on TensorFlow 1.\nThe latest release supporting TensorFlow 1 is [v1.5.1](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Freleases\u002Ftag\u002Fv1.5.1).\n[Documentation](https:\u002F\u002Fgpflow.readthedocs.io\u002Fen\u002Fv1.5.1-docs\u002F) and\n[tutorials](https:\u002F\u002Fnbviewer.jupyter.org\u002Fgithub\u002FGPflow\u002FGPflow\u002Fblob\u002Fv1.5.1\u002Fdoc\u002Fsphinx\u002Fnotebooks\u002Fintro.ipynb)\nwill remain available.\n\n\n## Citing GPflow\n\nTo cite GPflow, please reference the [JMLR paper](http:\u002F\u002Fwww.jmlr.org\u002Fpapers\u002Fvolume18\u002F16-537\u002F16-537.pdf). Sample Bibtex is given below:\n```\n@ARTICLE{GPflow2017,\n  author = {Matthews, Alexander G. de G. and {van der Wilk}, Mark and Nickson, Tom and\n\tFujii, Keisuke. and {Boukouvalas}, Alexis and {Le{\\'o}n-Villagr{\\'a}}, Pablo and\n\tGhahramani, Zoubin and Hensman, James},\n    title = \"{{GP}flow: A {G}aussian process library using {T}ensor{F}low}\",\n  journal = {Journal of Machine Learning Research},\n  year    = {2017},\n  month = {apr},\n  volume  = {18},\n  number  = {40},\n  pages   = {1-6},\n  url     = {http:\u002F\u002Fjmlr.org\u002Fpapers\u002Fv18\u002F16-537.html}\n}\n```\n\nSince the publication of the GPflow paper, the software has been significantly extended\nwith the framework for interdomain approximations and multioutput priors. We review the\nframework and describe the design in an [arXiv paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2003.01115),\nwhich can be cited by users.\n```\n@article{GPflow2020multioutput,\n  author = {{van der Wilk}, Mark and Dutordoir, Vincent and John, ST and\n            Artemev, Artem and Adam, Vincent and Hensman, James},\n  title = {A Framework for Interdomain and Multioutput {G}aussian Processes},\n  year = {2020},\n  journal = {arXiv:2003.01115},\n  url = {https:\u002F\u002Farxiv.org\u002Fabs\u002F2003.01115}\n}\n```\n","\u003C\u002Fthink>\n\n\u003Cdiv style=\"text-align:center\">\n\u003Cimg width=\"500\" height=\"200\" src=\"https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fblob\u002Fdevelop\u002Fdoc\u002Fsphinx\u002F_static\u002Fgpflow_logo.svg\">\n\u003C\u002Fdiv>\n\n[![CircleCI](https:\u002F\u002Fcircleci.com\u002Fgh\u002FGPflow\u002FGPflow\u002Ftree\u002Fdevelop.svg?style=svg)](https:\u002F\u002Fcircleci.com\u002Fgh\u002FGPflow\u002FGPflow\u002Ftree\u002Fdevelop)\n[![Coverage Status](http:\u002F\u002Fcodecov.io\u002Fgithub\u002FGPflow\u002FGPflow\u002Fcoverage.svg?branch=master)](http:\u002F\u002Fcodecov.io\u002Fgithub\u002FGPflow\u002FGPflow?branch=master)\n[![Slack Status](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fslack-gpflow-green.svg?logo=Slack)](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fgpflow\u002Fshared_invite\u002FenQtOTE5MDA0Nzg5NjA2LTYwZWI3MzhjYjNlZWI1MWExYzZjMGNhOWIwZWMzMGY0YjVkYzAyYjQ4NjgzNDUyZTgyNzcwYjAyY2QzMWRmYjE)\n\n\n[网站](https:\u002F\u002Fgpflow.org) |\n[文档（发布版）](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002F) |\n[文档（开发版）](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop) |\n[术语表](GLOSSARY.md)\n\n#### 目录\n\u003C!-- created with help from https:\u002F\u002Fgithub.com\u002Fekalinin\u002Fgithub-markdown-toc and further manual adjustments -->\n\n* [GPflow 是做什么的？](#what-does-gpflow-do)\n* [安装](#installation)\n* [开始使用 GPflow 2.0](#getting-started-with-gpflow-20)\n* [GPflow 社区](#the-gpflow-community)\n   * [获取帮助](#getting-help)\n   * [Slack 工作区](#slack-workspace)\n   * [贡献](#contributing)\n   * [使用 GPflow 的项目](#projects-using-gpflow)\n* [版本兼容性](#version-compatibility)\n   * [TensorFlow 1.x 和 GPflow 1.x](#tensorflow-1x-and-gpflow-1x)\n* [引用 GPflow](#citing-gpflow)\n\n\n## GPflow 是做什么的？\n\nGPflow 是一个用于在 Python 中构建高斯过程 (Gaussian process) 模型的包。\n它实现了适用于可组合核函数 (kernels) 和似然函数 (likelihoods) 的现代高斯过程推断 (inference)。\n\nGPflow 基于 [TensorFlow 2.4+](http:\u002F\u002Fwww.tensorflow.org) 和 [TensorFlow Probability](https:\u002F\u002Fwww.tensorflow.org\u002Fprobability\u002F) 运行计算，这使得在 GPU (图形处理器) 上快速执行成为可能。\n\n[在线文档（最新发行版）](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002F)\u002F[(开发版)](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop) 包含更多详细信息。\n\n\n### 维护者\n\n它最初由 [James Hensman](http:\u002F\u002Fjameshensman.github.io\u002F) 和 [Alexander G. de G. Matthews](https:\u002F\u002Fgithub.com\u002Falexggmatthews) 创建。\n现在由以下人员积极维护（按字母顺序排列）：\n[Artem Artemev](http:\u002F\u002Fgithub.com\u002Fawav\u002F)、\n[Mark van der Wilk](https:\u002F\u002Fmarkvdw.github.io\u002F)、\n[ST John](https:\u002F\u002Fgithub.com\u002Fst--)\n以及 [Vincent Dutordoir](https:\u002F\u002Fvdutor.github.io\u002F)。\n没有社区就没有现在的 GPflow。**我们要感谢所有为塑造 GPflow 做出贡献的 [贡献者](CONTRIBUTORS.md)。**\n\n *GPflow 是一个开源项目。如果您具备相关技能并有兴趣参与贡献，请联系我们（见下文“[GPflow 社区](#the-gpflow-community)\"部分）。*\n\n\n## 安装\n\n### 依赖要求\n\nGPflow 同时依赖于 TensorFlow (TF, 版本 ≥ 2.4) 和 TensorFlow Probability (TFP, 版本 ≥ 0.12)。我们支持 Python ≥ 3.7。\n\n**注意：** TensorFlow Probability 的发布与 TensorFlow 紧密耦合，例如 TFP 0.14 需要 TF>=2.6，TFP 0.13 需要 TF>=2.5，而 TFP 0.12 需要 TF>=2.4。不幸的是，这一点并未在 TFP 的依赖项中指定。因此，如果您已经安装了（旧版）TensorFlow，GPflow 将拉取最新的 TFP，这可能会导致不兼容。如果您遇到类似 `ImportError: This version of TensorFlow Probability requires TensorFlow version >= 2.4` 的错误，您必须升级 TensorFlow (`pip install -U tensorflow`) 或手动安装较旧版本的 `tensorflow_probability` 包。\n\n### 来自 PyPI 的最新（稳定）发行版\n\n```bash\npip install gpflow\n```\n\n### 来自 GitHub 的最新（前沿）源码\n\n*请注意，`develop` 分支可能会经常更改，新的提交可能会破坏您的代码。*\n\n在 [GPflow GitHub 仓库](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow) 的 `develop` 分支检出后，运行\n```bash\npip install -e .\n```\n\n或者，您可以使用 `pip` 安装最新的 GitHub `develop` 版本：\n```bash\npip install git+https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow.git@develop#egg=gpflow\n```\n这将自动安装所有必需的依赖项。\n\n## 开始使用 GPflow 2.0\n\n有一个 [\"Intro to GPflow 2.0\"](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop\u002Fnotebooks\u002Fintro_to_gpflow2.html) Jupyter notebook (Jupyter 笔记本)；请查看它以了解详细信息。\n要将您的代码从 GPflow 1 转换过来，请查看 [GPflow 2 升级指南](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop\u002Fnotebooks\u002Fgpflow2_upgrade_guide.html)。\n\n\n## GPflow 社区\n\n### 获取帮助\n\n**Bug、功能请求、痛点、令人烦恼的设计怪癖等：**\n请使用 [GitHub issues](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fissues\u002F) (问题追踪系统) 来标记 bug\u002F问题\u002F痛点，建议新功能，并讨论任何与 GPflow 使用相关的其他内容，这些内容在某种意义上涉及更改 GPflow 代码本身。\n您可以使用 [标签](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels)，例如 [`bug`](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels\u002Fbug)、[`discussion`](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels\u002Fdiscussion)、[`feature`](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels\u002Ffeature)、[`feedback`](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Flabels\u002Ffeedback) 等。\n我们非常欢迎关于可用性的评论或担忧，以及任何设计层面的变更建议。\n\n我们的目标是及时响应问题，但如果您认为我们可能忘记了某个问题，请随时添加另一个评论提醒我们。\n\n**“如何使用”类问题：**\n请使用 [Stack Overflow (gpflow 标签)](https:\u002F\u002Fstackoverflow.com\u002Ftags\u002Fgpflow) (开发者问答社区) 询问与“如何使用 GPflow\"相关的问题，即理解性问题，而不是需要更改 GPflow 代码的问题。（如果您不确定在哪里提问，始终欢迎打开一个 GitHub issue；我们随后可能会要求您将问题移至 Stack Overflow。）\n\n### Slack 工作区\n\n我们有一个公共的 [GPflow slack 工作区](https:\u002F\u002Fgpflow.slack.com\u002F) (即时通讯平台)。如果您想加入，无论是为了提出简短的非正式问题，还是为了参与 GPflow 的讨论和未来开发，请使用此 [邀请链接](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fgpflow\u002Fshared_invite\u002FenQtOTE5MDA0Nzg5NjA2LTYwZWI3MzhjYjNlZWI1MWExYzZjMGNhOWIwZWMzMGY0YjVkYzAyYjQ4NjgzNDUyZTgyNzcwYjAyY2QzMWRmYjE)。\n\n### 贡献\n\n我们非常感谢所有的建设性意见。更多信息，请参阅 [贡献者指南](CONTRIBUTING.md)。\n\n### 使用 GPflow 的项目\n\n以下列出了基于 GPflow 构建并展示其用法的项目。以下项目均基于当前的 GPflow 2.x 版本：\n\n| 项目 | 描述 |\n| --- | --- |\n| [Trieste](https:\u002F\u002Fgithub.com\u002Fsecondmind-labs\u002Ftrieste)       | 使用 TensorFlow 进行贝叶斯优化，开箱即用支持 GPflow (2.x) 模型。 |\n| [VFF](https:\u002F\u002Fgithub.com\u002Fst--\u002FVFF)       | 用于高斯过程 (Gaussian Processes) 的变分傅里叶特征 (GPflow 2.x 版本) |\n| [BranchedGP](https:\u002F\u002Fgithub.com\u002FManchesterBioinference\u002FBranchedGP) | 具有分支核的高斯过程。|\n| [VBPP](https:\u002F\u002Fgithub.com\u002Fst--\u002Fvbpp) | “点过程变分贝叶斯”的实现。|\n| [Gaussian Process Regression on Molecules](https:\u002F\u002Fmedium.com\u002F@ryangriff123\u002Fgaussian-process-regression-on-molecules-in-gpflow-ee6fedab2130) | 通过创建自定义定义的 Tanimoto 核来处理 Morgan 指纹，从而预测分子属性的高斯过程 (GPs) |\n\n如果您希望您的项目列在这里，请告诉我们——或者直接 [提交一个拉取请求](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fcompare)，将您的项目添加到上面的表格中！\n\n*以下项目基于 GPflow 的旧版本（2020 年之前）；我们鼓励其作者升级到 GPflow 2。*\n\n| 项目 | 描述 |\n| --- | --- |\n| [GPflowOpt](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflowOpt)       | 使用 GPflow 进行贝叶斯优化（稳定版需要 GPflow 0.5）。 |\n| [Doubly-Stochastic-DGP](https:\u002F\u002Fgithub.com\u002FICL-SML\u002FDoubly-Stochastic-DGP)| 具有双重随机变分推断的深度高斯过程。|\n| [widedeepnetworks](https:\u002F\u002Fgithub.com\u002Fwidedeepnetworks\u002Fwidedeepnetworks) | 测量随机宽深度神经网络与高斯过程 (GPs) 之间的关系。|\n| [orth_decoupled_var_gps](https:\u002F\u002Fgithub.com\u002Fhughsalimbeni\u002Forth_decoupled_var_gps) | 具有正交解耦基底的变分稀疏高斯过程 (GPs)|\n| [kernel_learning](https:\u002F\u002Fgithub.com\u002Ffrgsimpson\u002Fkernel_learning) | “高斯过程可微组合核学习”的实现。|\n| [DGPs_with_IWVI](https:\u002F\u002Fgithub.com\u002Fhughsalimbeni\u002FDGPs_with_IWVI) | 具有重要性加权变分推断的深度高斯过程|\n| [kerndisc](https:\u002F\u002Fgithub.com\u002FBracketJohn\u002FkernDisc) | 用于单变量数据自动核结构发现的库|\n| [Signature covariances](https:\u002F\u002Fgithub.com\u002Ftgcsaba\u002FGPSig) | 将 (时间) 序列作为*输入*的核函数 |\n| [Structured-DGP](https:\u002F\u002Fgithub.com\u002Fboschresearch\u002FStructured_DGP) | 为双重随机深度高斯过程的变分后验添加更多结构 |\n\n## 版本兼容性\n\nGPflow 严重依赖 TensorFlow，只要 TensorFlow 支持前向兼容性，GPflow 也应该如此。GPflow 的版本号可为您提供有关向后兼容性的线索。如果主版本号发生了变化，则需要查看发布说明以了解 API 是如何更改的。\n\n不幸的是，GPflow _模型_ 不存在向后兼容性的概念，这意味着模型实现可以在不改变接口的情况下发生变化。换句话说，对于来自不同版本 GPflow 的相同模型，TensorFlow 计算图可能会有所不同。\n\n### TensorFlow 1.x 和 GPflow 1.x\n\n我们已经停止了对基于 TensorFlow 1 的 GPflow 的开发和支持。支持 TensorFlow 1 的最新版本是 [v1.5.1](https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Freleases\u002Ftag\u002Fv1.5.1)。[文档](https:\u002F\u002Fgpflow.readthedocs.io\u002Fen\u002Fv1.5.1-docs\u002F) 和\n[教程](https:\u002F\u002Fnbviewer.jupyter.org\u002Fgithub\u002FGPflow\u002FGPflow\u002Fblob\u002Fv1.5.1\u002Fdoc\u002Fsphinx\u002Fnotebooks\u002Fintro.ipynb)\n仍将可用。\n\n\n## 引用 GPflow\n\n要引用 GPflow，请参考 [JMLR 论文](http:\u002F\u002Fwww.jmlr.org\u002Fpapers\u002Fvolume18\u002F16-537\u002F16-537.pdf)。以下是示例 BibTeX：\n```\n@ARTICLE{GPflow2017,\n  author = {Matthews, Alexander G. de G. and {van der Wilk}, Mark and Nickson, Tom and\n\tFujii, Keisuke. and {Boukouvalas}, Alexis and {Le{\\'o}n-Villagr{\\'a}}, Pablo and\n\tGhahramani, Zoubin and Hensman, James},\n    title = \"{{GP}flow: A {G}aussian process library using {T}ensor{F}low}\",\n  journal = {Journal of Machine Learning Research},\n  year    = {2017},\n  month = {apr},\n  volume  = {18},\n  number  = {40},\n  pages   = {1-6},\n  url     = {http:\u002F\u002Fjmlr.org\u002Fpapers\u002Fv18\u002F16-537.html}\n}\n```\n\n自 GPflow 论文发表以来，该软件已通过域间近似和多输出先验框架得到了显著扩展。我们在 [arXiv 论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2003.01115) 中回顾了该框架并描述了设计，用户可以进行引用。\n```\n@article{GPflow2020multioutput,\n  author = {{van der Wilk}, Mark and Dutordoir, Vincent and John, ST and\n            Artemev, Artem and Adam, Vincent and Hensman, James},\n  title = {A Framework for Interdomain and Multioutput {G}aussian Processes},\n  year = {2020},\n  journal = {arXiv:2003.01115},\n  url = {https:\u002F\u002Farxiv.org\u002Fabs\u002F2003.01115}\n}\n```","# GPflow 快速上手指南\n\nGPflow 是一个用于构建高斯过程（Gaussian Process）模型的 Python 包。它基于 [TensorFlow](http:\u002F\u002Fwww.tensorflow.org) 和 [TensorFlow Probability](https:\u002F\u002Fwww.tensorflow.org\u002Fprobability\u002F) 实现现代高斯过程推断，支持 GPU 加速计算。\n\n## 1. 环境准备\n\n在开始之前，请确保您的开发环境满足以下要求：\n\n- **Python 版本**: ≥ 3.7\n- **TensorFlow (TF)**: ≥ 2.4\n- **TensorFlow Probability (TFP)**: ≥ 0.12\n\n> **注意**：TensorFlow Probability 的版本与 TensorFlow 紧密耦合（例如 TFP 0.12 需要 TF>=2.4）。如果您已安装旧版 TensorFlow，直接安装 GPflow 可能会因依赖冲突报错。如遇 `ImportError`，请先升级 TensorFlow (`pip install -U tensorflow`) 或手动指定兼容的 TFP 版本。\n\n## 2. 安装步骤\n\n推荐使用国内镜像源以加快下载速度。\n\n### 安装稳定版（推荐）\n```bash\npip install -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple gpflow\n```\n\n### 安装最新开发版（Develop Branch）\n*警告：`develop` 分支可能频繁变动，新提交可能导致代码不兼容。*\n\n```bash\npip install git+https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow.git@develop#egg=gpflow\n```\n\n## 3. 基本使用\n\nGPflow 的核心功能通过 Jupyter Notebook 进行演示。获取完整示例的最佳方式是查阅官方入门教程。\n\n### 最小化代码示例\n```python\nimport gpflow\nimport tensorflow as tf\n\n# 参考官方 Notebook 进行模型定义与训练\n# 链接：https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop\u002Fnotebooks\u002Fintro_to_gpflow2.html\n```\n\n### 下一步操作\n1. 访问 **[Intro to GPflow 2.0 Notebook]**(https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop\u002Fnotebooks\u002Fintro_to_gpflow2.html) 查看详细代码示例。\n2. 如果您是从 GPflow 1.x 迁移，请参考 **[升级指南]**(https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Fdevelop\u002Fnotebooks\u002Fgpflow2_upgrade_guide.html)。\n3. 遇到问题可查阅 [在线文档](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002F) 或在 [Stack Overflow](https:\u002F\u002Fstackoverflow.com\u002Ftags\u002Fgpflow) 提问。","某工业物联网团队负责预测精密设备的剩余使用寿命，需要在数据稀缺且噪声较大的情况下准确评估预测结果的风险边界。\n\n### 没有 GPflow 时\n- 必须手写复杂的矩阵运算和协方差函数，不仅开发周期长，还极易引入数值稳定性问题导致模型崩溃\n- 传统实现仅支持 CPU 计算，处理大规模时序数据时训练速度极慢，完全无法满足生产环境的实时性要求\n- 无法灵活地将深度学习特征嵌入概率框架，模型泛化能力受限，难以应对复杂工况\n- 缺乏原生的不确定性量化接口，业务方难以理解预测结果的置信程度，影响决策信任度\n\n### 使用 GPflow 后\n- 利用现成的可组合核函数库，几天内即可搭建起完整的回归预测原型，大幅降低开发门槛\n- 依托 TensorFlow 后端自动调度 GPU 资源，千级样本的训练时间从数小时缩短至秒级，显著提升迭代效率\n- 无缝集成 TensorFlow Probability，方便构建混合了神经网络的深度高斯过程模型，捕捉更复杂的非线性关系\n- 直接获取预测分布的均值与方差，为设备维护决策提供明确的概率依据，增强系统可靠性\n\nGPflow 通过简化贝叶斯推断的工程实现，帮助团队在有限数据下高效交付可靠的预测系统。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FGPflow_GPflow_75907477.png","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FGPflow_151b8569.png","",null,"gpflow.org","https:\u002F\u002Fgithub.com\u002FGPflow",[81,85],{"name":82,"color":83,"percentage":84},"Python","#3572A5",99.8,{"name":86,"color":87,"percentage":88},"Makefile","#427819",0.2,1906,430,"2026-04-02T08:33:19","Apache-2.0","未说明","非必需，支持 GPU 加速，具体型号、显存及 CUDA 版本未说明",{"notes":96,"python":97,"dependencies":98},"TensorFlow Probability 版本与 TensorFlow 版本紧密耦合（如 TFP 0.14 需 TF>=2.6），安装时需注意兼容性；若遇到 ImportError 可能需要手动指定旧版 TFP 或升级 TF；GPflow 1.x 已停止维护，建议使用 2.x 版本并参考升级指南。","3.7+",[99,100],"tensorflow>=2.4","tensorflow_probability>=0.12",[15,26,13],[103,104,105,106,107,108,109,110,111,112,113],"gaussian-processes","tensorflow","gpflow","machine-learning","variational-inference","bayesian-statistics","markov-chain-monte-carlo","stochastic-processes","deep-learning","ml","gp","2026-03-27T02:49:30.150509","2026-04-06T05:15:47.780299",[117,122,126,131,136,141],{"id":118,"question_zh":119,"answer_zh":120,"source_url":121},3348,"如何在 GPFlow 中实现结合 TensorFlow 神经网络的端到端训练？","可以使用 SVGP 模型连接神经网络最后一层和 GP。通过优化 GP 模型的 objective 函数来同时更新 GP 参数和 NN 权重。示例代码：`gp_model = gpf.models.SVGP(h, phs.label, kernel, likelihood, ...)`，然后 `minimise = optimiser.minimize(gp_model.objective)`。确保将网络最后隐藏层作为输入传递给 GP 模型，这样梯度下降可以同时优化两者。","https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fissues\u002F505",{"id":123,"question_zh":124,"answer_zh":125,"source_url":121},3349,"训练过程中测试集似然值停滞不前怎么办？","这可能与随机初始化有关。如果在大多数情况下表现合理但偶尔停滞，可能是初始化导致陷入局部最优。建议检查代码逻辑是否等价于分别训练两个网络再连接，或尝试不同的初始化策略。参考相关讨论中的调试经验，确认目标函数是否正确传递了所有 Trainable variables。",{"id":127,"question_zh":128,"answer_zh":129,"source_url":130},3350,"为什么最近版本的 GPFlow 在 GPU 上运行速度比 CPU 慢？","这是由于引入了批量矩阵三角求解（batch matrix triangular solve）和 VecToTri\u002FTriToVec 操作导致的性能回归。这些新操作在 GPU 上的并行化效果不如旧版，且内存操作开销大。建议检查 TensorFlow 版本兼容性，或使用 profiling 工具（如 chrome:\u002F\u002Ftracing）分析具体瓶颈。","https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fissues\u002F181",{"id":132,"question_zh":133,"answer_zh":134,"source_url":135},3351,"如何处理具有不同测量不确定性的数据（向量方差）？","标准高斯似然通常假设恒定不确定性。对于动态不确定性，可参考 Issue #619 的解释。或者考虑使用 SVGP\u002FVGP 模型处理长序列平滑任务，它们支持更灵活的噪声建模。对于 2000+ 点的序列，SVGP 是推荐方案。","https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fissues\u002F550",{"id":137,"question_zh":138,"answer_zh":139,"source_url":140},3352,"使用 Matern 核进行回归时报错，RBF 核却正常，原因是什么？","这通常是数值稳定性问题。Matern 核在优化过程中的 Cholesky 分解可能比 RBF 核更容易出现数值误差。如果遇到此类错误，建议优先使用 RBF 核，或检查数据预处理及超参数范围。TensorFlow 编译时的指令集警告也可能影响性能但不一定是报错主因。","https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fissues\u002F490",{"id":142,"question_zh":143,"answer_zh":144,"source_url":145},3353,"GPU 上的矩阵三角求解是否需要 CUSolver？","不需要。当前实现仅调用 CUBLAS 而不需要 CUSOLVER。移除这一瓶颈可以将 SVGP 的 GPU 性能提升约 3 倍。Cholesky 分解才可能需要 CUSOLVER。查看相关 fork 确认 GPU 矩阵三角求解实现只调用了 CUBLAS。","https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fissues\u002F213",[147,152,157,161,166,171,176,181,186,191,196,201,206,211,216,221,226,231,236,241],{"id":148,"version":149,"summary_zh":150,"released_at":151},102928,"v2.8.0","The main focus of this release is to provide users control over arguments for `tf.function`\r\ncompilation inside the Scipy minimize wrapper. It also adds support for a new categorical kernel.\r\n\r\n## Major Features and Improvements\r\n\r\n* Added a new categorical kernel that implements categorical variables by mapping them to values in\r\n  a latent space. (#2055)\r\n* Added support for passing `tf.function` arguments for compilation in `gpflow.optimizers.Scipy`.\r\n  (#2064)\r\n* Default lower bound for parameters of scalar likelihoods can now be set via configuration.\r\n  (#1985, #2066)\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Fixed some notebook typos and a link. (#2052, #2057)\r\n* Fixed missing docs for `SquaredExponential` and `Constant` kernels. (#2056, #2063)\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nsc336, partev, khurram-ghani, uri-granta, awav, jesnie","2023-05-03T22:33:00",{"id":153,"version":154,"summary_zh":155,"released_at":156},102929,"v2.6.5","A small fix for a bug in the scipy optimize wrapper.\r\n\r\n## Breaking Changes\r\n\r\n* None\r\n\r\n## Known Caveats\r\n\r\n* None\r\n\r\n## Major Features and Improvements\r\n\r\n* None\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Patched `gpflow.optimizers.Scipy` to always assign the last good state returned by `scipy.optimize.minimize` to the model under optimization.  Previously, this step could be missed if `minimize` failed in some situations, leaving the model in an arbitrary state.\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\nkhurram-ghani\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FGPflow\u002FGPflow\u002Fcompare\u002Fv2.6.4...v2.6.5","2023-03-16T15:27:20",{"id":158,"version":159,"summary_zh":155,"released_at":160},102930,"v2.7.1","2023-03-16T08:57:51",{"id":162,"version":163,"summary_zh":164,"released_at":165},102931,"v2.7.0","The main theme of this release is documentation, with a new suite of tutorials, several upgrades to notebooks and the removal of a rather annoying bug in the documentation site.\r\n\r\nPerhaps more notably, `check_shapes` has been removed, and can now be found [here](https:\u002F\u002Fgithub.com\u002FGPflow\u002Fcheck_shapes).  This change is breaking for those who are still getting `check_shapes` from `gpflow`, although being in experimental this change does not require a new version number.\r\n\r\n## Breaking Changes\r\n\r\n* `gpflow.experimental.check_shapes` has been removed, in favour of an independent release. Use\r\n  `pip install check_shapes` and `import check_shapes` instead.\r\n\r\n## Major Features and Improvements\r\n\r\n* Major rework of documentation landing page and \"getting started\" section.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Fixed bug related to `tf.saved_model` and methods wrapped in `@check_shapes`.\r\n* Documented monitoring with `Adam` optimizer.\r\n* Fixed bug related to switching versions in documentation site\r\n* Fixed several issues relating to mypy\r\n\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nsc336, st--, sethaxen, jesnie\r\n","2023-01-27T11:59:40",{"id":167,"version":168,"summary_zh":169,"released_at":170},102932,"v2.6.4","This is yet another bug-fix release.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Fix to `to_default_float` to avoid losing precision when called with python floats.\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nChrisMorter\r\n","2022-12-02T15:24:19",{"id":172,"version":173,"summary_zh":174,"released_at":175},102933,"v2.6.3","# Release 2.6.3 (next upcoming release in progress)\r\n\r\nThis is yet another bug-fix release.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Fix to `check_shapes` handling of `tfp..._TensorCoercible`.\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\njesnie","2022-10-13T09:20:05",{"id":177,"version":178,"summary_zh":179,"released_at":180},102934,"v2.6.2","# Release 2.6.2\r\n\r\nThis is a bug-fix release, for compatibility with GPflux.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Extract shapes of `tfp.python.layers.internal.distribution_tensor_coercible._TensorCoercible`.\r\n* Allow `FallbackSeparateIndependentInducingVariables` to have children with different shapes.\r\n* Allow input and output batches on `GaussianQuadrature` to be different.\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\njesnie","2022-10-10T12:56:39",{"id":182,"version":183,"summary_zh":184,"released_at":185},102935,"v2.6.1","# Release 2.6.1\r\n\r\nThis is a bug-fixes release, due to problems with model saving in `2.6.0`.\r\n\r\n## Breaking Changes\r\n\r\n* Removed `gpflow.utilities.ops.cast`. Use `tf.cast` instead.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Fixed bug related to `tf.saved_model` and methods wrapped in `@check_shapes`.\r\n* Some documentation formatting fixes.\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\njesnie","2022-09-29T15:32:25",{"id":187,"version":188,"summary_zh":189,"released_at":190},102936,"v2.6.0","# Release 2.6.0\r\n\r\nThe major theme for this release is heteroskedastic likelihoods. Changes have unfortunately caused some breaking changes, but makes it much easier to use heteroskedastic likelihoods, either by plugging together built-in GPflow classes, or when writing your own. See our [updated notebook](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002F2.6.0\u002Fnotebooks\u002Fadvanced\u002Fvarying_noise.html), for examples on how to use this.\r\n\r\n## Breaking Changes\r\n\r\n* All likelihood methods now take an extra `X` argument. If you have written custom likelihoods or you have custom code calling likelihoods directly you will need to add this extra argument.\r\n* On the `CGLB` model the `xnew` parameters has changed name to `Xnew`, to be consistent with the other models.\r\n* On the `GPLVM` model the variance returned by `predict_f` with `full_cov=True` has changed shape from `[batch..., N, N, P]` to `[batch..., P, N, N]` to be consistent with the other models.\r\n* `gpflow.likelihoods.Gaussian.DEFAULT_VARIANCE_LOWER_BOUND` has been replaced with `gpflow.likelihoods.scalar_continuous.DEFAULT_LOWER_BOUND`.\r\n* Change to `InducingVariables` API. `InducingVariables` must now have a `shape` property.\r\n* `gpflow.experimental.check_shapes.get_shape.register` has been replaced with `gpflow.experimental.check_shapes.register_get_shape`.\r\n* `check_shapes` will no longer automatically wrap shape checking in `tf.compat.v1.flags.tf_decorator.make_decorator`. This is likely to affect you if you use `check_shapes` with custom Keras models. If you require the decorator you can manually enable it with `check_shapes(..., tf_decorator=True)`.\r\n\r\n## Known Caveats\r\n\r\n* Shape checking is now, by default, disabled within `tf.function`. Use `set_enable_check_shapes` to change this behaviour. See the [API documentation](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002F2.6.0\u002Fapi\u002Fgpflow\u002Fexperimental\u002Fcheck_shapes\u002Findex.html#speed-and-interactions-with-tf-function) for more details.\r\n\r\n## Major Features and Improvements\r\n\r\n* Improved handling of variable noise\r\n  - All likelihood methods now take an `X` argument, allowing you to easily implement heteroskedastic likelihoods.\r\n  - The `Gaussian` likelihood can now be parametrized by either a `variance` or a `scale`\r\n  - Some existing likelihoods can now take a function (of X) instead of a parameter, allowing them to become heteroskedastic. The parameters are:\r\n    - `Gaussian` `variance`\r\n    - `Gaussian` `scale`\r\n    - `StudentT` `scale`\r\n    - `Gamma` `shape`\r\n    - `Beta` `scale`\r\n  - The `GPR` and `SGPR` can now be configured with a custom Gaussian likelihood, allowing you to make them heteroskedastic.\r\n  - See the updated [notebook](https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002F2.6.0\u002Fnotebooks\u002Fadvanced\u002Fvarying_noise.html).\r\n  - `gpflow.mean_functions` has been renamed `gpflow.functions`, but with an alias, to avoid breaking changes.\r\n* `gpflow.experimental.check_shapes`\r\n  - Can now be in three different states - ENABLED, EAGER_MODE_ONLY, and DISABLE. The default is EAGER_MODE_ONLY, which only performs shape checks when the code is not compiled. Compiling the shape checking code is a major bottleneck and this provides a significant speed-up for performance sensitive parts of the code.\r\n  - Now supports multiple variable-rank dimensions at the same time, e.g. `cov: [n..., n...]`.\r\n  - Now supports single broadcast dimensions to have size 0 or 1, instead of only 1.\r\n  - Now supports variable-rank dimensions to be broadcast, even if they're not leading.\r\n  - Now supports `is None` and `is not None` as checks for conditional shapes.\r\n  - Now uses custom function `register_get_shape` instead of `get_shape.register`, for better compatibility with TensorFlow.\r\n  - Now supports checking the shapes of `InducingVariable`s.\r\n  - Now adds documentation to function arguments that has declared shapes, but no other documentation.\r\n  - All of GPflow is now consistently shape-checked.\r\n* All built-in kernels now consistently support broadcasting.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Tested with TensorFlow 2.10.\r\n* Add support for Apple Silicon Macs (`arm64`) via the `tensorflow-macos` dependency. (#1850)\r\n* New implementation of GPR and SGPR posterior objects. This primarily improves numerical stability. (#1960)\r\n  - For the GPR this is also a speed improvement when using a GPU.\r\n  - For the SGPR this is a mixed bag, performance-wise.\r\n* Improved checking and error reporting for the models than do not support `full_cov` and `full_output_cov`.\r\n* Documentation improvements:\r\n  - Improved MCMC notebook.\r\n  - Deleted notebooks that had no contents.\r\n  - Fixed some broken formatting.\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\njesnie, corwinpro, st--, vdutor\r\n","2022-09-20T13:46:48",{"id":192,"version":193,"summary_zh":194,"released_at":195},102937,"v2.5.2","# Release 2.5.2\r\n\r\nThis release fixes a performance regression introduced in `2.5.0`.  `2.5.0` used features of Python\r\nthat `tensorfow \u003C 2.9.0` do not know how to compile, which negatively impacted performance.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Fixed some bugs that prevented TensorFlow compilation and had negative performance impact. (#1882)\r\n* Various improvements to documentation. (#1875, #1866, #1877, #1879)\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\njesnie","2022-05-10T10:29:01",{"id":197,"version":198,"summary_zh":199,"released_at":200},102938,"v2.5.1","# Release 2.5.1\r\n\r\nFix problem with release process of 2.5.0.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Fix bug in release process.\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\njesnie\r\n","2022-04-28T12:02:50",{"id":202,"version":203,"summary_zh":204,"released_at":205},102923,"v2.10.0","# Release 2.10.0\r\n\r\nThis release adds a utility function for converting tf.Modules to float32.\r\n\r\n## Improvements\r\n\r\n* Add the `freeze_as_float32` utility function, which returns a frozen deepcopy of a module with all values converted to tf.float32.\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nuri-granta\r\n","2025-05-29T13:02:38",{"id":207,"version":208,"summary_zh":209,"released_at":210},102924,"v2.9.2","This patch release adds support for Tensorflow 2.16.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Support and test against Tensorflow 2.16. **Note that (like tensorflow-probability) GPflow uses Keras 2. Since TF 2.16 defaults to Keras 3, `tf.keras` must now be imported from the `tf_keras` package. Alternatively, you can import `tf_keras` from the `gpflow.keras` module, which will automatically select the right source depending on which version of TF is installed.** Note that Keras optimizers such as `Adam` should be imported from `tf_keras`.\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nuri-granta\r\n","2024-06-14T10:30:50",{"id":212,"version":213,"summary_zh":214,"released_at":215},102925,"v2.9.1","This patch release fixes a number of identified issues.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Support pickling Scipy optimizers with a non-empty compile cache\r\n* Allow setting a prior for A in the `Linear` mean function\r\n* Avoid rounding small values in kernel summary printout\r\n* Test against Tensorflow 2.15\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nJSchmiegel, uri-granta\r\n","2024-02-07T14:27:37",{"id":217,"version":218,"summary_zh":219,"released_at":220},102926,"v2.9.0","# Release 2.9.0\r\n\r\nThis release adds caching of compiled graphs inside the Scipy optimizer, and adds support\r\nfor returning loss history. It also adds supports for Python 3.11.\r\n\r\n## Major Features and Improvements\r\n\r\n* Support returning loss history with Scipy optimizer. \r\n* Scipy minimize wrapper caches compiled graphs and re-uses them if called with the same arguments.\r\n  This functionality can be disabled by setting the new `compile_cache_size` argument to 0.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Support and test with Python 3.11\r\n* Test against a 'production' environment (in addition to 'min' and 'max' environments).\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nkhurram-ghani, jesnie, uri-granta\r\n","2023-08-09T08:36:14",{"id":222,"version":223,"summary_zh":224,"released_at":225},102927,"v2.8.1","A small fix to ensure support for (and testing with) TensorFlow 2.12.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Support and test with TensorFlow 2.12\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nuri-granta\r\n","2023-06-27T12:15:00",{"id":227,"version":228,"summary_zh":229,"released_at":230},102939,"v2.5.0","# Release 2.5.0\r\n\r\nThe focus of this release has mostly been bumping the minimally supported versions of Python and\r\nTensorFlow; and development of `gpflow.experimental.check_shapes`.\r\n\r\n## Breaking Changes\r\n\r\n* Dropped support for Python 3.6. New minimum version is 3.7. (#1803, #1859)\r\n* Dropped support for TensorFlow 2.2 and 2.3. New minimum version is 2.4. (#1803)\r\n* Removed sub-package `gpflow.utilities.utilities`. It was scheduled for deletion in `2.3.0`.\r\n  Use `gpflow.utilities` instead. (#1804)\r\n* Removed method `Likelihood.predict_density`, which has been deprecated since March 24, 2020.\r\n  (#1804)\r\n* Removed property `ScalarLikelihood.num_gauss_hermite_points`, which has been deprecated since\r\n  September 30, 2020. (#1804)\r\n\r\n## Known Caveats\r\n\r\n* Further improvements to type hints - this may reveal new problems in your code-base if\r\n  you use a type checker, such as `mypy`. (#1795, #1799, #1802, #1812, #1814, #1816)\r\n\r\n## Major Features and Improvements\r\n\r\n* Significant work on `gpflow.experimental.check_shapes`.\r\n\r\n  - Support anonymous dimensions. (#1796)\r\n  - Add a hook to let the user register shapes for custom types. (#1798)\r\n  - Support `Optional` values. (#1797)\r\n  - Make it configurable. (#1810)\r\n  - Add accesors for setting\u002Fgetting previously applied checks. (#1815)\r\n  - Much improved error messages. (#1822)\r\n  - Add support for user notes on shapes. (#1836)\r\n  - Support checking all elements of collections. (#1840)\r\n  - Enable stand-alone shape checking, without using a decorator. (#1845)\r\n  - Support for broadcasts. (#1849)\r\n  - Add support for checking the shapes of intermediate computations. (#1853)\r\n  - Support conditional shapes. (#1855)\r\n\r\n* Significant speed-up of the GPR posterior objects. (#1809, #1811)\r\n\r\n* Significant improvements to documentation. Note the new home page:\r\n  https:\u002F\u002Fgpflow.github.io\u002FGPflow\u002Findex.html\r\n  (#1828, #1829, #1830, #1831, #1833, #1841, #1842, #1856, #1857)\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Minor improvement to code clarity (variable scoping) in SVGP model. (#1800)\r\n* Improving mathematical formatting in docs (SGPR derivations). (#1806)\r\n* Allow anisotropic kernels to have negative length-scales. (#1843)\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nltiao, uri.granta, frgsimpson, st--, jesnie","2022-04-28T10:27:15",{"id":232,"version":233,"summary_zh":234,"released_at":235},102940,"v2.4.0","# Release 2.4.0\r\n\r\nThis release mostly focuses on make posterior objects useful for Bayesian Optimisation.\r\nIt also adds a new `experimetal` sub-package, with a tool for annotating tensor shapes.\r\n\r\n\r\n## Breaking Changes\r\n\r\n* Slight change to the API of custom posterior objects.\r\n  `gpflow.posteriors.AbstractPosterior._precompute` no longer must return an `alpha` and an\r\n  `Qinv` - instead it returns any arbitrary tuple of `PrecomputedValue`s.\r\n  Correspondingly `gpflow.posteriors.AbstractPosterior._conditional_with_precompute` should no\r\n  longer try to access `self.alpha` and `self.Qinv`, but instead is passed the tuple of tensors\r\n  returned by `_precompute`, as a parameter. (#1763, #1767)\r\n\r\n* Slight change to the API of inducing points.\r\n  You should no longer override `gpflow.inducing_variables.InducingVariables.__len__`. Override\r\n  `gpflow.inducing_variables.InducingVariables.num_inducing` instead. `num_inducing` should return a\r\n  `tf.Tensor` which is consistent with previous behaviour, although the type previously was\r\n  annotated as `int`. `__len__` has been deprecated. (#1766, #1792)\r\n\r\n## Known Caveats\r\n\r\n* Type hints have been added in several places - this may reveal new problems in your code-base if\r\n  you use a type checker, such as `mypy`.\r\n  (#1766, #1769, #1771, #1773, #1775, #1777, #1780, #1783, #1787, #1789)\r\n\r\n## Major Features and Improvements\r\n\r\n* Add new posterior class to enable faster predictions from the VGP model. (#1761)\r\n* VGP class bug-fixed to work with variable-sized data. Note you can use\r\n  `gpflow.models.vgp.update_vgp_data` to ensure variational parameters are updated sanely. (#1774).\r\n* All posterior classes bug-fixed to work with variable data sizes, for Bayesian Optimisation.\r\n  (#1767)\r\n\r\n* Added `experimental` sub-package for features that are still under developmet.\r\n  * Added `gpflow.experimental.check_shapes` for checking tensor shapes.\r\n    (#1760, #1768, #1782, #1785, #1788)\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Make `dataclasses` dependency conditional at install time. (#1759)\r\n* Simplify calculations of some `predict_f`. (#1755)\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\njesnie, tmct, joacorapela\r\n","2022-03-01T12:39:12",{"id":237,"version":238,"summary_zh":239,"released_at":240},102941,"v2.3.1","# Release 2.3.1\r\n\r\nThis is a bug-fix release, primarily for the GPR posterior object.\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* GPR posterior\r\n  * Fix the calculation in the GPR posterior object (#1734).\r\n  * Fixes leading dimension issues with `GPRPosterior._conditional_with_precompute()` (#1747).\r\n\r\n* Make `gpflow.optimizers.Scipy` able to handle unused \u002F unconnected variables. (#1745).\r\n\r\n* Build\r\n  * Fixed broken CircleCi build (#1738).\r\n  * Update CircleCi build to use next-gen Docker images (#1740).\r\n  * Fixed broken triggering of docs generation (#1744).\r\n  * Make all slow tests depend on fast tests (#1743).\r\n  * Make `make dev-install` also install the test requirements (#1737).\r\n\r\n* Documentation\r\n  * Fixed broken link in `README.md` (#1736).\r\n  * Fix broken build of `cglb.ipynb` (#1742).\r\n  * Add explanation of how to run notebooks locally (#1729).\r\n  * Fix formatting in notebook on Heteroskedastic Likelihood (#1727).\r\n  * Fix broken link in introduction (#1718).\r\n\r\n* Test suite\r\n  * Amends `test_gpr_posterior.py` so it will cover leading dimension uses.\r\n\r\n\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\nst--, jesnie, johnamcleod, Andrew878","2022-01-20T11:31:52",{"id":242,"version":243,"summary_zh":244,"released_at":245},102942,"v2.3.0","## Major Features and Improvements\r\n\r\n* Refactor posterior base class to support other model types. (#1695)\r\n* Add new posterior class to enable faster predictions from the GPR\u002FSGPR models. (#1696, #1711)\r\n* Construct Parameters from other Parameters and retain properties. (#1699)\r\n* Add CGLB model (#1706)\r\n\r\n## Bug Fixes and Other Changes\r\n\r\n* Fix unit test failure when using TensorFlow 2.5.0 (#1684)\r\n* Upgrade black formatter to version 20.8b1 (#1694)\r\n* Remove erroneous DeprecationWarnings (#1693)\r\n* Fix SGPR derivation (#1688)\r\n* Fix tests which fail with TensorFlow 2.6.0 (#1714)\r\n\r\n## Thanks to our Contributors\r\n\r\nThis release contains contributions from:\r\n\r\njohnamcleod, st--, Andrew878, tadejkrivec, awav, avullo","2021-10-26T09:33:33"]