[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-Nixtla--mlforecast":3,"tool-Nixtla--mlforecast":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",154349,2,"2026-04-13T23:32:16",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"status":17},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",108322,"2026-04-10T11:39:34",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":32,"last_commit_at":59,"category_tags":60,"status":17},4721,"markitdown","microsoft\u002Fmarkitdown","MarkItDown 是一款由微软 AutoGen 团队打造的轻量级 Python 工具，专为将各类文件高效转换为 Markdown 格式而设计。它支持 PDF、Word、Excel、PPT、图片（含 OCR）、音频（含语音转录）、HTML 乃至 YouTube 链接等多种格式的解析，能够精准提取文档中的标题、列表、表格和链接等关键结构信息。\n\n在人工智能应用日益普及的今天，大语言模型（LLM）虽擅长处理文本，却难以直接读取复杂的二进制办公文档。MarkItDown 恰好解决了这一痛点，它将非结构化或半结构化的文件转化为模型“原生理解”且 Token 效率极高的 Markdown 格式，成为连接本地文件与 AI 分析 pipeline 的理想桥梁。此外，它还提供了 MCP（模型上下文协议）服务器，可无缝集成到 Claude Desktop 等 LLM 应用中。\n\n这款工具特别适合开发者、数据科学家及 AI 研究人员使用，尤其是那些需要构建文档检索增强生成（RAG）系统、进行批量文本分析或希望让 AI 助手直接“阅读”本地文件的用户。虽然生成的内容也具备一定可读性，但其核心优势在于为机器",93400,"2026-04-06T19:52:38",[52,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":66,"readme_en":67,"readme_zh":68,"quickstart_zh":69,"use_case_zh":70,"hero_image_url":71,"owner_login":72,"owner_name":72,"owner_avatar_url":73,"owner_bio":74,"owner_company":75,"owner_location":75,"owner_email":76,"owner_twitter":77,"owner_website":78,"owner_url":79,"languages":80,"stars":96,"forks":97,"last_commit_at":98,"license":99,"difficulty_score":100,"env_os":101,"env_gpu":102,"env_ram":102,"env_deps":103,"category_tags":116,"github_topics":117,"view_count":32,"oss_zip_url":75,"oss_zip_packed_at":75,"status":17,"created_at":124,"updated_at":125,"faqs":126,"releases":156},7374,"Nixtla\u002Fmlforecast","mlforecast","Scalable machine 🤖 learning for time series forecasting.","mlforecast 是一个专为时间序列预测设计的机器学习框架，旨在帮助用户高效、准确地利用海量数据训练预测模型。它主要解决了现有 Python 工具在处理大规模时间序列时速度慢、精度低且难以扩展的痛点，让机器学习模型能够轻松应用于生产环境。\n\n无论是需要处理百万级时间序列的数据科学家、机器学习工程师，还是从事量化分析的研究人员，都能通过 mlforecast 快速构建高性能的预测方案。其核心亮点在于拥有目前 Python 生态中最快的时间序列特征工程实现，并完美兼容 pandas、polars、spark、dask 和 ray 等多种数据处理后端，支持分布式计算以应对超大数据量。\n\n此外，mlforecast 采用了类似 scikit-learn 的简洁语法（.fit 和 .predict），降低了学习门槛，同时支持外生变量、静态协变量以及基于共形预测的概率 forecasting 功能。如果你正在寻找一个既能保证速度又能灵活扩展的时间序列预测工具，mlforecast 将是一个非常务实的选择。","# mlforecast\n\n[![Tweet](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Furl\u002Fhttp\u002Fshields.io.svg?style=social)](https:\u002F\u002Ftwitter.com\u002Fintent\u002Ftweet?text=Statistical%20Forecasting%20Algorithms%20by%20Nixtla%20&url=https:\u002F\u002Fgithub.com\u002FNixtla\u002Fstatsforecast&via=nixtlainc&hashtags=StatisticalModels,TimeSeries,Forecasting)\n[![Slack](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSlack-4A154B?&logo=slack&logoColor=white.png)](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fnixtlacommunity\u002Fshared_invite\u002Fzt-1pmhan9j5-F54XR20edHk0UtYAPcW4KQ)\n\n\n\u003Cdiv align=\"center\">\n\u003Ccenter>\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FNixtla_mlforecast_readme_0eeb1aa47fc5.png\" \u002F>\u003C\u002Fcenter>\n\u003Ch1 align=\"center\">Machine Learning 🤖 Forecast\u003C\u002Fh1>\n\u003Ch3 align=\"center\">Scalable machine learning for time series forecasting\u003C\u002Fh3>\n\n[![CI](https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Factions\u002Fworkflows\u002Fci.yaml\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Factions\u002Fworkflows\u002Fci.yaml)\n[![Python](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fpyversions\u002Fmlforecast.png)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fmlforecast\u002F)\n[![PyPi](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fmlforecast?color=blue.png)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fmlforecast\u002F)\n[![conda-forge](https:\u002F\u002Fimg.shields.io\u002Fconda\u002Fvn\u002Fconda-forge\u002Fmlforecast?color=blue.png)](https:\u002F\u002Fanaconda.org\u002Fconda-forge\u002Fmlforecast)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flicense\u002FNixtla\u002Fmlforecast.png)](https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Fblob\u002Fmain\u002FLICENSE)\n\n**mlforecast** is a framework to perform time series forecasting using\nmachine learning models, with the option to scale to massive amounts of\ndata using remote clusters.\n\n\u003C\u002Fdiv>\n\n## Install\n\n### PyPI\n\n`pip install mlforecast`\n\n### conda-forge\n\n`conda install -c conda-forge mlforecast`\n\nFor more detailed instructions you can refer to the [installation\npage](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Finstall.html).\n\n## Quick Start\n\n1. **Get Started with this [quick\nguide](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Fquick_start_local.html).**\n2. **Follow this [end-to-end\nwalkthrough](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Fend_to_end_walkthrough.html)\nfor best practices.**\n\n### Videos\n\n- [Overview](https:\u002F\u002Fwww.youtube.com\u002Flive\u002FEnhyJx8l2LE)\n\n### Sample notebooks\n\n- [m5](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fm5-mlforecast-eval)\n- [m5-polars](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fm5-mlforecast-eval-polars)\n- [m4](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fm4-competition)\n- [m4-cv](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fm4-competition-cv)\n- [favorita](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fmlforecast-favorita)\n- [VN1](https:\u002F\u002Fcolab.research.google.com\u002Fdrive\u002F1UdhCAk49k6HgMezG-U_1ETnAB5pYvZk9)\n\n## Why?\n\nCurrent Python alternatives for machine learning models are slow,\ninaccurate and don’t scale well. So we created a library that can be\nused to forecast in production environments.\n[`MLForecast`](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fforecast.html#mlforecast)\nincludes efficient feature engineering to train any machine learning\nmodel (with `fit` and `predict` methods such as\n[`sklearn`](https:\u002F\u002Fscikit-learn.org\u002Fstable\u002F)) to fit millions of time\nseries.\n\n## Features\n\n- Fastest implementations of feature engineering for time series\n  forecasting in Python.\n- Out-of-the-box compatibility with pandas, polars, spark, dask, and\n  ray.\n- Probabilistic Forecasting with Conformal Prediction.\n- Support for exogenous variables and static covariates.\n- Familiar `sklearn` syntax: `.fit` and `.predict`.\n\nMissing something? Please open an issue or write us in\n[![Slack](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSlack-4A154B?&logo=slack&logoColor=white.png)](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fnixtlaworkspace\u002Fshared_invite\u002Fzt-135dssye9-fWTzMpv2WBthq8NK0Yvu6A)\n\n## Examples and Guides\n\n📚 [End to End\nWalkthrough](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Fend_to_end_walkthrough.html):\nmodel training, evaluation and selection for multiple time series.\n\n🔎 [Probabilistic\nForecasting](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Ftutorials\u002Fprediction_intervals_in_forecasting_models.html):\nuse Conformal Prediction to produce prediciton intervals.\n\n👩‍🔬 [Cross\nValidation](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fhow-to-guides\u002Fcross_validation.html):\nrobust model’s performance evaluation.\n\n🔁 [M5: Reuse CV Splits + Global\u002FGrouped Rolling Means](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fhow-to-guides\u002Fhyperparameter_optimization):\noptimize with cached CV windows while tuning global and grouped rolling features in one workflow.\n\n🔌 [Predict Demand\nPeaks](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Ftutorials\u002Felectricity_peak_forecasting.html):\nelectricity load forecasting for detecting daily peaks and reducing\nelectric bills.\n\n📈 [Transfer\nLearning](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fhow-to-guides\u002Ftransfer_learning.html):\npretrain a model using a set of time series and then predict another one\nusing that pretrained model.\n\n🌡️ [Distributed\nTraining](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Fquick_start_distributed.html):\nuse a Dask, Ray or Spark cluster to train models at scale.\n\n## How to use\n\nThe following provides a very basic overview, for a more detailed\ndescription see the\n[documentation](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002F).\n\n### Data setup\n\nStore your time series in a pandas dataframe in long format, that is,\neach row represents an observation for a specific serie and timestamp.\n\n``` python\nfrom mlforecast.utils import generate_daily_series\n\nseries = generate_daily_series(\n    n_series=20,\n    max_length=100,\n    n_static_features=1,\n    static_as_categorical=False,\n    with_trend=True\n)\nseries.head()\n```\n\n|     | unique_id | ds         | y          | static_0 |\n|-----|-----------|------------|------------|----------|\n| 0   | id_00     | 2000-01-01 | 17.519167  | 72       |\n| 1   | id_00     | 2000-01-02 | 87.799695  | 72       |\n| 2   | id_00     | 2000-01-03 | 177.442975 | 72       |\n| 3   | id_00     | 2000-01-04 | 232.704110 | 72       |\n| 4   | id_00     | 2000-01-05 | 317.510474 | 72       |\n\n\n> Note: The unique_id serves as an identifier for each distinct time\n> series in your dataset. If you are using only single time series from\n> your dataset, set this column to a constant value.\n\n### Models\n\nNext define your models, each one will be trained on all series. These\ncan be any regressor that follows the scikit-learn API.\n\n``` python\nimport lightgbm as lgb\nfrom sklearn.linear_model import LinearRegression\n```\n\n``` python\nmodels = [\n    lgb.LGBMRegressor(random_state=0, verbosity=-1),\n    LinearRegression(),\n]\n```\n\n### Forecast object\n\nNow instantiate an\n[`MLForecast`](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fforecast.html#mlforecast)\nobject with the models and the features that you want to use. The\nfeatures can be lags, transformations on the lags and date features. You\ncan also define transformations to apply to the target before fitting,\nwhich will be restored when predicting.\n\n``` python\nfrom mlforecast import MLForecast\nfrom mlforecast.lag_transforms import ExpandingMean, RollingMean\nfrom mlforecast.target_transforms import Differences\n```\n\n``` python\nfcst = MLForecast(\n    models=models,\n    freq='D',\n    lags=[7, 14],\n    lag_transforms={\n        1: [ExpandingMean()],\n        7: [RollingMean(window_size=28)]\n    },\n    date_features=['dayofweek'],\n    target_transforms=[Differences([1])],\n)\n```\n\n### Training\n\nTo compute the features and train the models call `fit` on your\n`Forecast` object.\n\n``` python\nfcst.fit(series)\n```\n\n```\nMLForecast(models=[LGBMRegressor, LinearRegression], freq=D, lag_features=['lag7', 'lag14', 'expanding_mean_lag1', 'rolling_mean_lag7_window_size28'], date_features=['dayofweek'], num_threads=1)\n```\n\n### Predicting\n\nTo get the forecasts for the next `n` days call `predict(n)` on the\nforecast object. This will automatically handle the updates required by\nthe features using a recursive strategy.\n\n``` python\npredictions = fcst.predict(14)\npredictions\n```\n\n\n|     | unique_id | ds         | LGBMRegressor | LinearRegression |\n|-----|-----------|------------|---------------|------------------|\n| 0   | id_00     | 2000-04-04 | 299.923771    | 311.432371       |\n| 1   | id_00     | 2000-04-05 | 365.424147    | 379.466214       |\n| 2   | id_00     | 2000-04-06 | 432.562441    | 460.234028       |\n| 3   | id_00     | 2000-04-07 | 495.628000    | 524.278924       |\n| 4   | id_00     | 2000-04-08 | 60.786223     | 79.828767        |\n| ... | ...       | ...        | ...           | ...              |\n| 275 | id_19     | 2000-03-23 | 36.266780     | 28.333215        |\n| 276 | id_19     | 2000-03-24 | 44.370984     | 33.368228        |\n| 277 | id_19     | 2000-03-25 | 50.746222     | 38.613001        |\n| 278 | id_19     | 2000-03-26 | 58.906524     | 43.447398        |\n| 279 | id_19     | 2000-03-27 | 63.073949     | 48.666783        |\n\n\u003Cp>280 rows × 4 columns\u003C\u002Fp>\n\n\n### Visualize results\n\n``` python\nfrom utilsforecast.plotting import plot_series\n```\n\n``` python\nfig = plot_series(series, predictions, max_ids=4, plot_random=False)\n```\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FNixtla_mlforecast_readme_2f09b4affef8.png)\n\n## How to contribute\n\nSee\n[CONTRIBUTING.md](https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Fblob\u002Fmain\u002FCONTRIBUTING.md).\n","# mlforecast\n\n[![Tweet](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Furl\u002Fhttp\u002Fshields.io.svg?style=social)](https:\u002F\u002Ftwitter.com\u002Fintent\u002Ftweet?text=Statistical%20Forecasting%20Algorithms%20by%20Nixtla%20&url=https:\u002F\u002Fgithub.com\u002FNixtla\u002Fstatsforecast&via=nixtlainc&hashtags=StatisticalModels,TimeSeries,Forecasting)\n[![Slack](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSlack-4A154B?&logo=slack&logoColor=white.png)](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fnixtlacommunity\u002Fshared_invite\u002Fzt-1pmhan9j5-F54XR20edHk0UtYAPcW4KQ)\n\n\n\u003Cdiv align=\"center\">\n\u003Ccenter>\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FNixtla_mlforecast_readme_0eeb1aa47fc5.png\" \u002F>\u003C\u002Fcenter>\n\u003Ch1 align=\"center\">机器学习 🤖 预测\u003C\u002Fh1>\n\u003Ch3 align=\"center\">面向时间序列预测的可扩展机器学习框架\u003C\u002Fh3>\n\n[![CI](https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Factions\u002Fworkflows\u002Fci.yaml\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Factions\u002Fworkflows\u002Fci.yaml)\n[![Python](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fpyversions\u002Fmlforecast.png)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fmlforecast\u002F)\n[![PyPi](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fmlforecast?color=blue.png)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fmlforecast\u002F)\n[![conda-forge](https:\u002F\u002Fimg.shields.io\u002Fconda\u002Fvn\u002Fconda-forge\u002Fmlforecast?color=blue.png)](https:\u002F\u002Fanaconda.org\u002Fconda-forge\u002Fmlforecast)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flicense\u002FNixtla\u002Fmlforecast.png)](https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Fblob\u002Fmain\u002FLICENSE)\n\n**mlforecast** 是一个用于基于机器学习模型进行时间序列预测的框架，支持通过远程集群扩展到海量数据。\n\n\u003C\u002Fdiv>\n\n## 安装\n\n### PyPI\n\n`pip install mlforecast`\n\n### conda-forge\n\n`conda install -c conda-forge mlforecast`\n\n更多详细说明请参阅[安装页面](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Finstall.html)。\n\n## 快速入门\n\n1. **从这份[快速指南](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Fquick_start_local.html)开始。**\n2. **遵循这份[端到端教程](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Fend_to_end_walkthrough.html)，了解最佳实践。**\n\n### 视频\n\n- [概述](https:\u002F\u002Fwww.youtube.com\u002Flive\u002FEnhyJx8l2LE)\n\n### 示例笔记本\n\n- [m5](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fm5-mlforecast-eval)\n- [m5-polars](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fm5-mlforecast-eval-polars)\n- [m4](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fm4-competition)\n- [m4-cv](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fm4-competition-cv)\n- [favorita](https:\u002F\u002Fwww.kaggle.com\u002Fcode\u002Flemuz90\u002Fmlforecast-favorita)\n- [VN1](https:\u002F\u002Fcolab.research.google.com\u002Fdrive\u002F1UdhCAk49k6HgMezG-U_1ETnAB5pYvZk9)\n\n## 为什么？\n\n目前用于机器学习模型的 Python 替代方案速度慢、精度低且难以扩展。因此，我们开发了一个可在生产环境中使用的库。[`MLForecast`](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fforecast.html#mlforecast) 包含高效的时间序列特征工程功能，可用于训练任何具有 `fit` 和 `predict` 方法的机器学习模型（例如 [`sklearn`](https:\u002F\u002Fscikit-learn.org\u002Fstable\u002F)），以适应数百万条时间序列。\n\n## 特性\n\n- Python 中最快的时间序列预测特征工程实现。\n- 开箱即用，兼容 pandas、polars、Spark、Dask 和 Ray。\n- 基于 Conformal Prediction 的概率预测。\n- 支持外生变量和静态协变量。\n- 熟悉的 `sklearn` 语法：`.fit` 和 `.predict`。\n\n缺少某些功能？请提交问题或在[Slack](https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fnixtlaworkspace\u002Fshared_invite\u002Fzt-135dssye9-fWTzMpv2WBthq8NK0Yvu6A) 上与我们交流。\n\n## 示例与指南\n\n📚 [端到端教程](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Fend_to_end_walkthrough.html)：多时间序列的模型训练、评估和选择。\n\n🔎 [概率预测](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Ftutorials\u002Fprediction_intervals_in_forecasting_models.html)：使用 Conformal Prediction 生成预测区间。\n\n👩‍🔬 [交叉验证](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fhow-to-guides\u002Fcross_validation.html)：稳健地评估模型性能。\n\n🔁 [M5：复用 CV 划分 + 全局\u002F分组滚动均值](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fhow-to-guides\u002Fhyperparameter_optimization.html)：在调优全局和分组滚动特征时，利用缓存的 CV 窗口优化模型。\n\n🔌 [预测需求高峰](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Ftutorials\u002Felectricity_peak_forecasting.html)：电力负荷预测，用于检测每日高峰并降低电费。\n\n📈 [迁移学习](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fhow-to-guides\u002Ftransfer_learning.html)：先用一组时间序列预训练模型，再用该预训练模型预测另一组数据。\n\n🌡️ [分布式训练](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Fquick_start_distributed.html)：使用 Dask、Ray 或 Spark 集群大规模训练模型。\n\n## 使用方法\n\n以下提供非常基础的概述，更详细的说明请参阅[文档](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002F)。\n\n### 数据准备\n\n将您的时间序列存储在一个长格式的 pandas 数据框中，即每行代表特定序列和时间戳的一个观测值。\n\n``` python\nfrom mlforecast.utils import generate_daily_series\n\nseries = generate_daily_series(\n    n_series=20,\n    max_length=100,\n    n_static_features=1,\n    static_as_categorical=False,\n    with_trend=True\n)\nseries.head()\n```\n\n|     | unique_id | ds         | y          | static_0 |\n|-----|-----------|------------|------------|----------|\n| 0   | id_00     | 2000-01-01 | 17.519167  | 72       |\n| 1   | id_00     | 2000-01-02 | 87.799695  | 72       |\n| 2   | id_00     | 2000-01-03 | 177.442975 | 72       |\n| 3   | id_00     | 2000-01-04 | 232.704110 | 72       |\n| 4   | id_00     | 2000-01-05 | 317.510474 | 72       |\n\n\n> 注意：unique_id 用作您数据集中每个独立时间序列的标识符。如果您只使用数据集中的单个时间序列，请将此列设置为常量值。\n\n### 模型\n\n接下来定义您的模型，每个模型都将针对所有序列进行训练。这些可以是任何遵循 scikit-learn API 的回归器。\n\n``` python\nimport lightgbm as lgb\nfrom sklearn.linear_model import LinearRegression\n```\n\n``` python\nmodels = [\n    lgb.LGBMRegressor(random_state=0, verbosity=-1),\n    LinearRegression(),\n]\n```\n\n### 预测对象\n\n现在实例化一个\n[`MLForecast`](https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fforecast.html#mlforecast)\n对象，传入你想要使用的模型和特征。这些特征可以是滞后项、对滞后项的变换以及日期特征。你还可以定义在拟合之前应用于目标变量的变换，这些变换会在预测时被还原。\n\n``` python\nfrom mlforecast import MLForecast\nfrom mlforecast.lag_transforms import ExpandingMean, RollingMean\nfrom mlforecast.target_transforms import Differences\n```\n\n``` python\nfcst = MLForecast(\n    models=models,\n    freq='D',\n    lags=[7, 14],\n    lag_transforms={\n        1: [ExpandingMean()],\n        7: [RollingMean(window_size=28)]\n    },\n    date_features=['dayofweek'],\n    target_transforms=[Differences([1])],\n)\n```\n\n### 训练\n\n要计算特征并训练模型，只需在你的 `Forecast` 对象上调用 `fit` 方法。\n\n``` python\nfcst.fit(series)\n```\n\n```\nMLForecast(models=[LGBMRegressor, LinearRegression], freq=D, lag_features=['lag7', 'lag14', 'expanding_mean_lag1', 'rolling_mean_lag7_window_size28'], date_features=['dayofweek'], num_threads=1)\n```\n\n### 预测\n\n要获取未来 `n` 天的预测结果，可以在预测对象上调用 `predict(n)` 方法。该方法会自动使用递归策略处理特征所需的更新。\n\n``` python\npredictions = fcst.predict(14)\npredictions\n```\n\n\n|     | unique_id | ds         | LGBMRegressor | LinearRegression |\n|-----|-----------|------------|---------------|------------------|\n| 0   | id_00     | 2000-04-04 | 299.923771    | 311.432371       |\n| 1   | id_00     | 2000-04-05 | 365.424147    | 379.466214       |\n| 2   | id_00     | 2000-04-06 | 432.562441    | 460.234028       |\n| 3   | id_00     | 2000-04-07 | 495.628000    | 524.278924       |\n| 4   | id_00     | 2000-04-08 | 60.786223     | 79.828767        |\n| ... | ...       | ...        | ...           | ...              |\n| 275 | id_19     | 2000-03-23 | 36.266780     | 28.333215        |\n| 276 | id_19     | 2000-03-24 | 44.370984     | 33.368228        |\n| 277 | id_19     | 2000-03-25 | 50.746222     | 38.613001        |\n| 278 | id_19     | 2000-03-26 | 58.906524     | 43.447398        |\n| 279 | id_19     | 2000-03-27 | 63.073949     | 48.666783        |\n\n\u003Cp>共280行 × 4列\u003C\u002Fp>\n\n\n### 可视化结果\n\n``` python\nfrom utilsforecast.plotting import plot_series\n```\n\n``` python\nfig = plot_series(series, predictions, max_ids=4, plot_random=False)\n```\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FNixtla_mlforecast_readme_2f09b4affef8.png)\n\n## 如何贡献\n\n请参阅\n[CONTRIBUTING.md](https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Fblob\u002Fmain\u002FCONTRIBUTING.md)。","# mlforecast 快速上手指南\n\n**mlforecast** 是一个专为时间序列预测设计的机器学习框架，支持高效特征工程，可轻松扩展至百万级时间序列，并兼容 pandas、polars、spark、dask 和 ray。\n\n## 环境准备\n\n- **操作系统**：Linux、macOS 或 Windows\n- **Python 版本**：3.8 及以上\n- **前置依赖**：无需特殊系统依赖，核心依赖包括 `scikit-learn`、`pandas`、`numpy` 等，安装时会自动解决。\n\n## 安装步骤\n\n推荐使用国内镜像源加速安装（如清华源）：\n\n```bash\npip install mlforecast -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n或使用 conda：\n\n```bash\nconda install -c conda-forge mlforecast\n```\n\n## 基本使用\n\n以下是最简使用示例，展示如何构建模型、训练并预测未来 14 天。\n\n### 1. 准备数据\n\n数据需为长格式（long format），包含 `unique_id`（序列 ID）、`ds`（时间戳）、`y`（目标值）。\n\n```python\nfrom mlforecast.utils import generate_daily_series\n\nseries = generate_daily_series(\n    n_series=20,\n    max_length=100,\n    n_static_features=1,\n    static_as_categorical=False,\n    with_trend=True\n)\nprint(series.head())\n```\n\n### 2. 定义模型\n\n可使用任意遵循 scikit-learn 接口的回归模型。\n\n```python\nimport lightgbm as lgb\nfrom sklearn.linear_model import LinearRegression\n\nmodels = [\n    lgb.LGBMRegressor(random_state=0, verbosity=-1),\n    LinearRegression(),\n]\n```\n\n### 3. 配置预测器\n\n设置频率、滞后特征、日期特征及目标变换。\n\n```python\nfrom mlforecast import MLForecast\nfrom mlforecast.lag_transforms import ExpandingMean, RollingMean\nfrom mlforecast.target_transforms import Differences\n\nfcst = MLForecast(\n    models=models,\n    freq='D',\n    lags=[7, 14],\n    lag_transforms={\n        1: [ExpandingMean()],\n        7: [RollingMean(window_size=28)]\n    },\n    date_features=['dayofweek'],\n    target_transforms=[Differences([1])],\n)\n```\n\n### 4. 训练模型\n\n```python\nfcst.fit(series)\n```\n\n### 5. 生成预测\n\n预测未来 14 天：\n\n```python\npredictions = fcst.predict(14)\nprint(predictions.head())\n```\n\n输出示例：\n\n|     | unique_id | ds         | LGBMRegressor | LinearRegression |\n|-----|-----------|------------|---------------|------------------|\n| 0   | id_00     | 2000-04-04 | 299.923771    | 311.432371       |\n| 1   | id_00     | 2000-04-05 | 365.424147    | 379.466214       |\n\n### 6. 可视化结果（可选）\n\n```python\nfrom utilsforecast.plotting import plot_series\n\nfig = plot_series(series, predictions, max_ids=4, plot_random=False)\n```\n\n至此，你已完成一个完整的时间序列预测流程。更多高级用法（如分布式训练、概率预测、交叉验证）请参考官方文档。","某大型连锁零售企业的数据团队需要为旗下 5000 家门店的 10 万种商品生成未来 30 天的销量预测，以优化库存周转。\n\n### 没有 mlforecast 时\n- **处理速度极慢**：传统 Python 循环方式处理百万级时间序列特征工程耗时数小时，无法支持每日频繁重训。\n- **扩展性差**：数据量增长后单机内存溢出，强行接入 Spark 或 Dask 需要重写大量底层代码，开发成本高昂。\n- **模型迭代困难**：难以快速对比 LightGBM、XGBoost 等不同算法在海量序列上的表现，导致只能沿用简单的统计模型，预测准确率偏低。\n- **缺乏不确定性评估**：仅能输出单一预测值，无法提供置信区间，业务部门不敢依据预测结果大胆调整备货策略。\n\n### 使用 mlforecast 后\n- **特征工程极速完成**：利用其优化的底层实现，几分钟内即可完成全量数据的滞后特征与滚动窗口计算，训练效率提升数十倍。\n- **无缝横向扩展**：凭借对 Polars、Spark 和 Ray 的原生支持，无需修改核心逻辑即可将任务分发至远程集群，轻松应对亿级数据规模。\n- **灵活模型集成**：通过类 sklearn 的统一接口（.fit\u002F.predict），快速遍历并部署多种机器学习模型，显著提升了长尾商品的预测精度。\n- **内置概率预测**：直接调用共形预测（Conformal Prediction）功能生成可靠的预测区间，帮助供应链团队制定更稳健的安全库存水位。\n\nmlforecast 将原本需要数天的大规模时序预测任务缩短至小时级，同时通过高精度的概率预测大幅降低了企业的库存积压与缺货风险。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FNixtla_mlforecast_2f09b4af.png","Nixtla","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FNixtla_3842e028.png","Open Source Time Series Ecosystem",null,"oss@nixtla.io","nixtlainc","www.nixtla.io","https:\u002F\u002Fgithub.com\u002FNixtla",[81,85,89,93],{"name":82,"color":83,"percentage":84},"Python","#3572A5",88.2,{"name":86,"color":87,"percentage":88},"Jupyter Notebook","#DA5B0B",11.4,{"name":90,"color":91,"percentage":92},"Shell","#89e051",0.2,{"name":94,"color":95,"percentage":92},"Makefile","#427819",1210,122,"2026-04-13T20:45:57","Apache-2.0",1,"Linux, macOS, Windows","未说明",{"notes":104,"python":105,"dependencies":106},"该工具是一个基于机器学习的时间序列预测框架，支持多种后端（pandas, polars, spark, dask, ray）以实现大规模数据扩展。模型需遵循 scikit-learn API（具备 fit 和 predict 方法）。可通过 PyPI 或 conda-forge 安装。文档提供了详细的快速入门和分布式训练指南。","3.8+",[107,108,109,110,111,112,113,114,115],"pandas","numpy","scikit-learn","lightgbm","polars","spark","dask","ray","utilsforecast",[14],[118,119,120,110,121,113,122,123],"forecast","forecasting","machine-learning","xgboost","python","time-series","2026-03-27T02:49:30.150509","2026-04-14T12:28:05.136481",[127,132,137,141,146,151],{"id":128,"question_zh":129,"answer_zh":130,"source_url":131},33099,"如何在使用 MLForecast 时正确处理周频（Weekly）数据的日期对齐问题？","周频数据常因聚合日期不一致导致报错。例如，你的数据是按周一聚合的，但 MLForecast 默认将周频（freq=\"W\"）数据聚合到周日。\n解决方法：\n1. 检查输入数据框中的日期是否是每周的同一天（如都是周一）。\n2. 将输入数据的日期调整为与模型频率一致的日期（如改为每周日）。\n3. 或者在初始化模型时明确指定频率对齐方式，确保训练数据和预测未来的日期索引完全匹配。","https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Fissues\u002F336",{"id":133,"question_zh":134,"answer_zh":135,"source_url":136},33094,"如何在 MLForecast 中保存和加载训练好的模型？","MLForecast 支持模型的保存和加载。如果在保存后加载模型无法预测结果，请检查以下几点：\n1. 确保使用的库版本是最新的，官方文档中有相关测试案例验证该功能：https:\u002F\u002Fnixtlaverse.nixtla.io\u002Fmlforecast\u002Fdocs\u002Fgetting-started\u002Fquick_start_distributed.html#saving-and-loading-2\n2. 确认保存和加载的代码逻辑正确，加载后应能直接调用 predict 方法。\n如果问题依旧，可能是环境或版本兼容性问题，建议升级相关依赖包。","https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Fissues\u002F288",{"id":138,"question_zh":139,"answer_zh":140,"source_url":131},33095,"遇到错误 'Found missing inputs in X_df' 该如何解决？","该错误通常是因为传入的 X_df（外生变量数据框）缺少预测 horizon 所需的某些时间戳或 ID 组合。解决方法包括：\n1. 使用 `MLForecast.make_future_dataframe(h)` 生成正确的未来数据结构作为参考。\n2. 使用 `MLForecast.get_missing_future(h, X_df)` 检查具体缺失了哪些行。\n3. 常见原因：日期频率不匹配。例如，数据按周一聚合，但模型频率设置为 \"W\"（默认聚合到周日），导致日期对不上。需统一输入数据和模型设置的日期对齐方式（如都改为周日）。\n4. 确保 pandas 版本较新（建议配合 utilsforecast>=0.1.5 使用），旧版本可能存在兼容性 bug。",{"id":142,"question_zh":143,"answer_zh":144,"source_url":145},33096,"如果时间序列数据中存在缺失的时间戳，应该如何处理以便进行预测？","如果训练集中缺少某些时间戳，会导致预测时出现 'Found missing inputs in X_df' 错误。解决方案是将时间列转换为整数索引：\n1. 使用以下代码将时间戳替换为连续的整数：\n```python\ndata['timestamp'] = data.sort_values(['unique_id', 'timestamp']).groupby('unique_id').cumcount()\n```\n2. 初始化模型时设置 `freq=1`：\n```python\nmodel = MLForecast(models=models, freq=1, lags=[12], ...)\n```\n注意：使用整数时间戳后将无法使用基于日期的特征（如 dayofweek, month），需移除 `date_features` 参数。","https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Fissues\u002F242",{"id":147,"question_zh":148,"answer_zh":149,"source_url":150},33097,"在直接预测模式（Direct Forecasts）下，外生变量（Exogenous Variables）是如何被处理的？","在直接预测模式下（每个 horizon 使用独立模型），目前实现可能存在局限性：即使设置了较长的预测 horizon（如 50），`predict` 方法可能仅使用了 X_df 中对应 horizon=1 的数据，而忽略了后续 horizon 对应的外生变量值（如傅里叶变换生成的季节性变量）。\n这意味着如果你传入的 X_df 中包含针对未来不同时间点的具体数值（如第 50 天的正弦\u002F余弦值），模型可能并未正确使用它们。\n建议：在使用直接预测法时，务必通过交叉验证测试外生变量是否真正影响了预测结果。如果发现行为不符合预期，可能需要等待官方修复或暂时使用递归预测模式。","https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Fissues\u002F496",{"id":152,"question_zh":153,"answer_zh":154,"source_url":155},33098,"DistributedMLForecast 的 predict 方法是否支持传入 X_df 参数？","在部分旧版本中，`DistributedMLForecast` 的 `predict` 方法可能不支持 `X_df` 关键字参数，从而抛出 `TypeError: predict() got an unexpected keyword argument 'X_df'` 错误。\n解决方案：\n1. 升级到最新版本的 mlforecast 库，新版本通常已修复此问题并支持分布式场景下的外生变量传入。\n2. 如果无法升级，请检查文档确认当前版本是否支持该功能，或尝试将外生变量合并到主数据框中再进行分布式预测。","https:\u002F\u002Fgithub.com\u002FNixtla\u002Fmlforecast\u002Fissues\u002F266",[157,162,167,172,177,182,187,192,197,202,207,212,217,222,227,232,237,242,247,252],{"id":158,"version":159,"summary_zh":160,"released_at":161},253688,"v1.0.31","本次发布不包含错误修复或新功能，仅将 `distributed` 模块添加到 `1.0.3` 版本中。\n\n## 变更\n- [修复] 在 wheel 构建中包含 distributed 模块 @nasaul (#592)\n- [修复] 跳过 dependabot PR 的文档构建 @nasaul (#589)\n- [文档] 更新仓库中的链接 @deven367 (#583)\n\n","2026-03-10T19:05:42",{"id":163,"version":164,"summary_zh":165,"released_at":166},253689,"v1.0.3","## 功能特性\n\n- [新功能] 为 MLForecast 添加数据验证 @nasaul (#560)\n- [新功能] 自动线程数设置 @nasaul (#559)\n- [新功能] 在分布式学习 API 中实现 sample_weight 的使用 @simonez-tuidi (#558)\n- [新功能] 实现针对特定个体预测范围的直接预测 @nasaul (#556)\n- [新功能] 在更高聚合级别上使用窗口函数 @janrth (#551)\n- [新功能] 对 update 方法进行增强和验证 @janrth (#541)\n- [新功能] 为 AutoSeasonalityAndDifferences 添加保护机制 @nasaul (#547)\n- [新功能] 在调参过程中可重复使用交叉验证划分 @janrth (#545)\n- [新功能] 在 light gbmcv 中添加特征\u002F权重列 @janrth (#532)\n- [新功能] 在 automl 中为 fit 方法添加 weight_col 参数 @janrth (#531)\n- [新功能] 增加在 fit 函数中传递额外参数的可能性 @PierD86 (#506)\n\n## 修复内容\n\n- [修复] 防止迁移学习预测后状态损坏 @W057 (#569)\n- [修复] 自动差分拟合后的逆变换 @janrth (#564)\n- [修复] 在交叉验证中使用样本权重 @nasaul (#555)\n- [修复] 实现直接预测的解决方案 @nasaul (#550)\n- [修复] 修复 Ray 相关问题并更新测试套件 @nasaul (#544)\n\n## 文档与杂项工作\n\n- [杂项] 发布准备 @nasaul (#575)\n- [杂项] 添加对 Python 3.14 的支持 @nasaul (#572)\n- [SEO] 更新标题 @deven367 (#567)\n- [杂项] 更新 uv.lock 中的 wheel 包 @nasaul (#553)\n- [杂项] 更新 Dask 版本以消除漏洞 @nasaul (#552)\n- [杂项] CI 更新 @nasaul (#549)\n- [杂项] 更新 CONTRIBUTING.md 文件 @nasaul (#546)\n- [文档] 修复新的增量预测示例渲染问题 @nasaul (#543)\n- [文档] 增量预测笔记本 @Satyajit-Chaudhuri (#526)\n- [杂项] 修复 Windows 测试套件问题 @nasaul (#542)\n- [杂项] 修复发布逻辑中的 bug @deven367 (#540)\n- [杂项] 启用工作流调度 @deven367 (#539)\n- [文档] 修正语法和拼写错误 @deven367 (#537)\n- [文档] 迁移文档 JSON 文件 @deven367 (#536)\n- [杂项] 从基于社区的 PR 的 build-docs 中移除部署步骤 @nasaul (#535)\n- [文档] 将 lazydocs 转换为 mkdocstrings @deven367 (#519)\n- [杂项] 更新 lightgbm 版本 @nasaul (#534)\n- [文档] 修复表格渲染问题 @deven367 (#525)\n- [杂项] 第三方许可证信息 @deven367 (#523)\n- [文档] 更多改进 @deven367 (#513)\n- [文档] 修复损坏的链接 @deven367 (#512)\n- [杂项] 设置 \"distributed.spark.lgb\" 环境变量 @deven367 (#510)\n- [文档] 更多文档修复 @deven367 (#509)\n- [文档] 安装用于 build-docs 的开发依赖项 @deven367 (#508)\n- [文档] 将测试从 nbs 迁移到 pytest @deven367 (#502)","2026-02-25T23:37:09",{"id":168,"version":169,"summary_zh":170,"released_at":171},253690,"v1.0.2","## 错误修复\n\n- 修复(compat)：处理 shift_array 中的零偏移 @jmoralez (#480)\n- 修复(global-sklearn-tfm)：对每一列应用逆变换 @jmoralez (#477)\n","2025-02-18T18:28:22",{"id":173,"version":174,"summary_zh":175,"released_at":176},253691,"v1.0.1","## Bug 修复\n\n- 修复：直接法中 X_df 的处理 @jmoralez (#468)\n- 修复（自动）：从 XGBoost 默认空间中移除无效参数 @jmoralez (#464)","2025-01-14T19:23:03",{"id":178,"version":179,"summary_zh":180,"released_at":181},253692,"v1.0.0","## 破坏性变更\n\n- 破坏性：移除 window_ops 和 numba 依赖 @jmoralez (#462)\n","2024-12-06T18:26:41",{"id":183,"version":184,"summary_zh":185,"released_at":186},253693,"v0.15.1","## 变更\n- chore: 弃用 window_ops @jmoralez (#410)\n\n## 新特性\n\n- feat(distributed): 在 predict 中支持 ID @jmoralez (#454)\n- feat(auto): 支持 input_size @jmoralez (#451)\n","2024-11-28T20:10:59",{"id":188,"version":189,"summary_zh":190,"released_at":191},253694,"v0.15.0","## 破坏性变更\n\n- 破坏性：当 `dropna=False` 时，删除目标列包含空值的行 @jmoralez (#447)\n\n## Bug 修复\n\n- 修复(auto)：支持自定义列名 @jmoralez (#449)\n\n## 功能增强\n\n- 增强(distributed)：在 Spark 中传播空特征 @jmoralez (#448)\n","2024-11-14T17:43:57",{"id":193,"version":194,"summary_zh":195,"released_at":196},253695,"v0.14.0","## 新特性\n\n- 功能：在 `MLForecast.fit` 和 `MLForecast.cross_validation` 中添加 `weight_col` 参数 @jmoralez (#444)\n- 功能：自动推断内置滞后变换所需的样本量 更新 @jmoralez (#445)","2024-11-11T19:23:47",{"id":198,"version":199,"summary_zh":200,"released_at":201},253696,"v0.13.6","## Bug 修复\n\n- 修复（分布式）：在分布式交叉验证中处理外生变量 @jmoralez (#443)\n- 修复（分布式）：支持预计算特征 @jmoralez (#436)","2024-11-08T18:01:59",{"id":203,"version":204,"summary_zh":205,"released_at":206},253697,"v0.13.5","## 增强功能\n\n- enh: 向 AutoMLForecast 添加 `step_size` 参数 @jmoralez (#426)\n- 在 `optimization.mlforecast_objective` 中支持 `step_size` 的选择 @bchaoss (#419)\n- 使用 TypeVar 定义 DataFrame 类型，并分发 `py.typed` 文件 @jmoralez (#408)\n","2024-10-10T21:14:49",{"id":208,"version":209,"summary_zh":210,"released_at":211},253698,"v0.13.4","## New Features\r\n\r\n- feat: mlflow flavor @jmoralez (#406)\r\n\r\n## Documentation\r\n\r\n- Clear up the README for the new user   @Ammar-Azman (#397)\r\n\r\n## Enhancement\r\n\r\n- make season_length optional in AutoMLForecast @jmoralez (#399)\r\n","2024-08-23T05:24:40",{"id":213,"version":214,"summary_zh":215,"released_at":216},253699,"v0.13.3","## Bug Fixes\r\n\r\n- handle no target transforms in DistributedMLForecast.to_local @jmoralez (#388)\r\n\r\n## Enhancement\r\n\r\n- ensure static features are constant @jmoralez (#391)\r\n","2024-07-25T18:33:42",{"id":218,"version":219,"summary_zh":220,"released_at":221},253700,"v0.13.2","## New Features\r\n\r\n- support prediction intervals in auto @jmoralez (#370)\r\n\r\n## Bug Fixes\r\n\r\n- remove dots from feature names in distributed @jmoralez (#382)\r\n- fix min_samples_split in random forest space @jmoralez (#380)\r\n\r\n## Enhancement\r\n\r\n- store prediction intervals inputs in MLForecast.save @jmoralez (#383)\r\n- support polars in GlobalSklearnTransformer @jmoralez (#377)\r\n","2024-07-17T19:27:31",{"id":223,"version":224,"summary_zh":225,"released_at":226},253701,"v0.13.1","## Dependencies\r\n\r\n- add polars extra @jmoralez (#368)\r\n- support polars 1.0 @jmoralez (#366)\r\n\r\n## Enhancement\r\n\r\n- add fitted argument to AutoMLForecast.fit @jmoralez (#351)\r\n","2024-07-01T18:38:47",{"id":228,"version":229,"summary_zh":230,"released_at":231},253702,"v0.13.0","## Breaking Change\r\n\r\n- set `refit=False` and `results_` as dict in AutoMLForecast @jmoralez (#341)\r\n\r\n## Bug fixes\r\n- fix: fitted nonrecursive cv with horizon >= 10 @adriaanvh1 (#333)\r\n\r\n## Enhancement\r\n\r\n- speedup date features @jmoralez (#340)\r\n- Create CODE_OF_CONDUCT.md @tracykteal (#335)\r\n","2024-05-09T18:38:56",{"id":233,"version":234,"summary_zh":235,"released_at":236},253703,"v0.12.1","## New Features\r\n\r\n- add auto module for hyperparameter optimization @tblume1992 (#306)\r\n- add DistributedMLForecast.update @jmoralez (#324)\r\n\r\n## Bug Fixes\r\n\r\n- fix cv fitted values with prediction intervals @jmoralez (#330)\r\n","2024-04-08T20:57:13",{"id":238,"version":239,"summary_zh":240,"released_at":241},253704,"v0.12.0","## Enhancement\r\n\r\n- migrate to coreforecast @jmoralez (#311)\r\n","2024-03-04T18:03:47",{"id":243,"version":244,"summary_zh":245,"released_at":246},253705,"v0.11.8","## Bug Fixes\r\n\r\n- ensure coreforecast is installed for AutoDifferences @jmoralez (#314)\r\n","2024-02-16T19:26:53",{"id":248,"version":249,"summary_zh":250,"released_at":251},253706,"v0.11.7","## New Features\r\n\r\n- add auto differences @jmoralez (#310)\r\n","2024-02-15T17:05:38",{"id":253,"version":254,"summary_zh":255,"released_at":256},253707,"v0.11.6","## New Features\r\n\r\n- add to_local method to distributed forecast @jmoralez (#302)\r\n- support saving and loading forecast objects @jmoralez (#301)\r\n","2024-01-19T02:22:50"]