[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tool-stanfordmlgroup--ngboost":3,"similar-stanfordmlgroup--ngboost":191},{"id":4,"github_repo":5,"name":6,"description_en":7,"description_zh":8,"ai_summary_zh":8,"readme_en":9,"readme_zh":10,"quickstart_zh":11,"use_case_zh":12,"hero_image_url":13,"owner_login":14,"owner_name":15,"owner_avatar_url":16,"owner_bio":17,"owner_company":18,"owner_location":18,"owner_email":18,"owner_twitter":18,"owner_website":19,"owner_url":20,"languages":21,"stars":38,"forks":39,"last_commit_at":40,"license":41,"difficulty_score":42,"env_os":43,"env_gpu":43,"env_ram":43,"env_deps":44,"category_tags":48,"github_topics":50,"view_count":56,"oss_zip_url":18,"oss_zip_packed_at":18,"status":57,"created_at":58,"updated_at":59,"faqs":60,"releases":95},2915,"stanfordmlgroup\u002Fngboost","ngboost","Natural Gradient Boosting for Probabilistic Prediction","ngboost 是一个基于 Python 的开源机器学习库，专为“概率预测”而设计。与传统模型仅输出单一预测值不同，ngboost 能够预测目标变量的完整概率分布，不仅告诉你会发生什么，还能量化预测的不确定性（例如置信区间）。\n\n它主要解决了现有梯度提升算法在处理不确定性估计时的局限性。在医疗诊断、金融风控等对风险敏感的场景中，仅仅知道预测结果是不够的，了解预测的可信度同样关键。ngboost 通过引入“自然梯度”技术，将概率分布的参数作为优化目标，实现了更稳定、更准确的概率建模。\n\n这款工具非常适合数据科学家、机器学习工程师以及科研人员使用。如果你习惯使用 Scikit-Learn，ngboost 的学习曲线会非常平缓，因为它完全兼容 Scikit-Learn 的接口风格。其独特的技术亮点在于高度的模块化：用户可以灵活选择基础学习器、概率分布类型以及评分规则，轻松定制适合特定任务的模型。无论是进行复杂的学术研究，还是构建需要风险评估的工业级应用，ngboost 都能提供强大且可扩展的支持。","# NGBoost: Natural Gradient Boosting for Probabilistic Prediction\n\n\u003Ch4 align=\"center\">\n\n![Python package](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fngboost)\n[![GitHub Repo Size](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Frepo-size\u002Fstanfordmlgroup\u002Fngboost?label=Repo+Size)](https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fgraphs\u002Fcontributors)\n[![Github License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache%202.0-blue.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FApache-2.0)\n[![Code style: black](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fcode%20style-black-000000.svg)](https:\u002F\u002Fgithub.com\u002Fpsf\u002Fblack)\n[![PyPI](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fngboost?logo=pypi&logoColor=white)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fngboost)\n[![PyPI Downloads](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fdm\u002Fngboost?logo=icloud&logoColor=white)](https:\u002F\u002Fpypistats.org\u002Fpackages\u002Fngboost)\n\n\u003C\u002Fh4>\n\nngboost is a Python library that implements Natural Gradient Boosting, as described in [\"NGBoost: Natural Gradient Boosting for Probabilistic Prediction\"](https:\u002F\u002Fstanfordmlgroup.github.io\u002Fprojects\u002Fngboost\u002F). It is built on top of [Scikit-Learn](https:\u002F\u002Fscikit-learn.org\u002Fstable\u002F), and is designed to be scalable and modular with respect to choice of proper scoring rule, distribution, and base learner. A didactic introduction to the methodology underlying NGBoost is available in this [slide deck](https:\u002F\u002Fdocs.google.com\u002Fpresentation\u002Fd\u002F1Tn23Su0ygR6z11jy3xVNiLGv0ggiUQue\u002Fedit?usp=share_link&ouid=102290675300480810195&rtpof=true&sd=true).\n\n## Installation\n\n```sh\nvia pip\n\npip install --upgrade ngboost\n\nvia conda-forge\n\nconda install -c conda-forge ngboost\n```\n\n## Usage\n\nProbabilistic regression example on the Boston housing dataset:\n\n```python\nfrom ngboost import NGBRegressor\n\nfrom sklearn.datasets import fetch_california_housing\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import mean_squared_error\n\n# Load California housing dataset\ncal = fetch_california_housing()\nX, Y = cal.data, cal.target\n\nX_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2)\n\nngb = NGBRegressor().fit(X_train, Y_train)\nY_preds = ngb.predict(X_test)\nY_dists = ngb.pred_dist(X_test)\n\n# test Mean Squared Error\ntest_MSE = mean_squared_error(Y_preds, Y_test)\nprint('Test MSE', test_MSE)\n\n# test Negative Log Likelihood\ntest_NLL = -Y_dists.logpdf(Y_test).mean()\nprint('Test NLL', test_NLL)\n```\n\nDetails on available distributions, scoring rules, learners, tuning, and model interpretation are available in our [user guide](https:\u002F\u002Fstanfordmlgroup.github.io\u002Fngboost\u002Fintro.html), which also includes numerous usage examples and information on how to add new distributions or scores to NGBoost.\n\n## License\n\n[Apache License 2.0](https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fblob\u002Fmaster\u002FLICENSE).\n\n## Reference\n\nTony Duan, Anand Avati, Daisy Yi Ding, Khanh K. Thai, Sanjay Basu, Andrew Y. Ng, Alejandro Schuler. 2019.\nNGBoost: Natural Gradient Boosting for Probabilistic Prediction.\n[arXiv](https:\u002F\u002Farxiv.org\u002Fabs\u002F1910.03225)\n","# NGBoost：用于概率预测的自然梯度提升\n\n\u003Ch4 align=\"center\">\n\n![Python 包](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fngboost)\n[![GitHub 仓库大小](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Frepo-size\u002Fstanfordmlgroup\u002Fngboost?label=Repo+Size)](https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fgraphs\u002Fcontributors)\n[![Github 许可证](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache%202.0-blue.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FApache-2.0)\n[![代码风格：black](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fcode%20style-black-000000.svg)](https:\u002F\u002Fgithub.com\u002Fpsf\u002Fblack)\n[![PyPI](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fngboost?logo=pypi&logoColor=white)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fngboost)\n[![PyPI 下载量](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fdm\u002Fngboost?logo=icloud&logoColor=white)](https:\u002F\u002Fpypistats.org\u002Fpackages\u002Fngboost)\n\n\u003C\u002Fh4>\n\nngboost 是一个 Python 库，实现了自然梯度提升方法，如论文 [\"NGBoost: Natural Gradient Boosting for Probabilistic Prediction\"](https:\u002F\u002Fstanfordmlgroup.github.io\u002Fprojects\u002Fngboost\u002F) 中所述。它基于 [Scikit-Learn](https:\u002F\u002Fscikit-learn.org\u002Fstable\u002F) 构建，旨在提供高度可扩展性和模块化设计，用户可以根据需求选择合适的评分规则、分布和基学习器。关于 NGBoost 方法论的通俗介绍，请参阅此 [幻灯片](https:\u002F\u002Fdocs.google.com\u002Fpresentation\u002Fd\u002F1Tn23Su0ygR6z11jy3xVNiLGv0ggiUQue\u002Fedit?usp=share_link&ouid=102290675300480810195&rtpof=true&sd=true)。\n\n## 安装\n\n```sh\n通过 pip\n\npip install --upgrade ngboost\n\n通过 conda-forge\n\nconda install -c conda-forge ngboost\n```\n\n## 使用\n\n在波士顿房价数据集上的概率回归示例：\n\n```python\nfrom ngboost import NGBRegressor\n\nfrom sklearn.datasets import fetch_california_housing\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import mean_squared_error\n\n# 加载加州房价数据集\ncal = fetch_california_housing()\nX, Y = cal.data, cal.target\n\nX_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2)\n\nngb = NGBRegressor().fit(X_train, Y_train)\nY_preds = ngb.predict(X_test)\nY_dists = ngb.pred_dist(X_test)\n\n# 测试均方误差\ntest_MSE = mean_squared_error(Y_preds, Y_test)\nprint('测试 MSE', test_MSE)\n\n# 测试负对数似然\ntest_NLL = -Y_dists.logpdf(Y_test).mean()\nprint('测试 NLL', test_NLL)\n```\n\n有关可用分布、评分规则、学习器、调参以及模型解释的详细信息，请参阅我们的 [用户指南](https:\u002F\u002Fstanfordmlgroup.github.io\u002Fngboost\u002Fintro.html)，其中还包含大量使用示例，以及如何向 NGBoost 添加新分布或新评分的信息。\n\n## 许可证\n\n[Apache License 2.0](https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fblob\u002Fmaster\u002FLICENSE)。\n\n## 参考文献\n\nTony Duan, Anand Avati, Daisy Yi Ding, Khanh K. Thai, Sanjay Basu, Andrew Y. Ng, Alejandro Schuler. 2019.\nNGBoost：用于概率预测的自然梯度提升。\n[arXiv](https:\u002F\u002Farxiv.org\u002Fabs\u002F1910.03225)","# NGBoost 快速上手指南\n\nNGBoost (Natural Gradient Boosting) 是一个用于**概率预测**的 Python 库。基于 Scikit-Learn 构建，它不仅输出预测值，还能输出预测值的概率分布（如不确定性区间），适用于回归和分类任务。\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*   **操作系统**：Linux, macOS 或 Windows\n*   **Python 版本**：推荐 Python 3.7 及以上版本\n*   **前置依赖**：\n    *   `scikit-learn` (核心依赖)\n    *   `numpy`\n    *   `scipy`\n    *   `lifelines` (部分分布需要)\n    *   `tqdm` (进度条显示)\n\n> **提示**：安装 NGBoost 时，pip 通常会自动解决上述依赖。如果您使用国内网络，建议在安装命令中指定清华或阿里镜像源以加速下载。\n\n## 2. 安装步骤\n\n您可以选择通过 `pip` 或 `conda` 进行安装。\n\n### 方式一：使用 pip 安装（推荐）\n\n**通用命令：**\n```sh\npip install --upgrade ngboost\n```\n\n**国内加速命令（推荐中国开发者使用）：**\n```sh\npip install --upgrade ngboost -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n### 方式二：使用 conda 安装\n\n如果您使用 Anaconda 或 Miniconda 环境：\n```sh\nconda install -c conda-forge ngboost\n```\n\n## 3. 基本使用\n\n以下是一个最简单的**概率回归**示例，使用加州房价数据集演示如何训练模型、获取点预测值以及预测分布。\n\n```python\nfrom ngboost import NGBRegressor\n\nfrom sklearn.datasets import fetch_california_housing\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import mean_squared_error\n\n# 加载加州房价数据集\ncal = fetch_california_housing()\nX, Y = cal.data, cal.target\n\n# 划分训练集和测试集\nX_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2)\n\n# 初始化并训练模型\nngb = NGBRegressor().fit(X_train, Y_train)\n\n# 获取点预测值 (Predicted Mean)\nY_preds = ngb.predict(X_test)\n\n# 获取预测分布对象 (Predicted Distribution)\nY_dists = ngb.pred_dist(X_test)\n\n# 评估指标 1: 均方误差 (MSE)\ntest_MSE = mean_squared_error(Y_preds, Y_test)\nprint('Test MSE', test_MSE)\n\n# 评估指标 2: 负对数似然 (Negative Log Likelihood, NLL)\n# 用于衡量概率预测的准确性\ntest_NLL = -Y_dists.logpdf(Y_test).mean()\nprint('Test NLL', test_NLL)\n```\n\n### 核心功能说明：\n*   `NGBRegressor()`: 用于回归任务的概率提升树模型。\n*   `.fit()`: 训练模型，接口与 Scikit-Learn 完全兼容。\n*   `.predict()`: 返回预测分布的均值（即传统回归模型的预测结果）。\n*   `.pred_dist()`: **核心特性**，返回完整的概率分布对象。您可以从中计算任意分位数、置信区间或特定值的概率密度。","某电商公司的数据科学团队正在构建房价预测模型，旨在为房产估值提供不仅包含预测值、还能量化不确定性风险的决策支持。\n\n### 没有 ngboost 时\n- **缺乏风险量化**：传统回归模型（如 XGBoost 或随机森林）仅输出单一预测值，业务方无法得知该预测值的置信区间，难以评估极端行情下的潜在亏损。\n- **分布假设僵化**：若强行使用统计方法估算误差，往往需预设数据服从正态分布，而实际房价数据常呈现偏态或厚尾特征，导致风险评估失真。\n- **校准成本高昂**：为了获得概率输出，团队需额外开发复杂的后处理脚本进行分位数回归或多模型集成，代码维护困难且计算资源消耗大。\n- **决策依据单一**：在制定信贷额度或库存策略时，因缺少对“预测可能性”的度量，只能依赖经验法则，错失优化机会。\n\n### 使用 ngboost 后\n- **原生概率预测**：ngboost 直接输出完整的概率分布（如对数正态分布），团队可瞬间获取任意置信水平的预测区间，清晰界定风险边界。\n- **灵活分布适配**：无需预设数据分布形态，ngboost 支持根据数据特性自动选择或自定义最匹配的分布函数，显著提升了对非典型房价数据的拟合精度。\n- **端到端高效建模**：基于自然梯度提升算法，ngboost 在 Scikit-Learn 生态中即可一键训练出概率模型，省去了繁琐的后处理步骤，开发效率提升数倍。\n- **精细化决策支持**：业务部门利用预测分布的方差信息，动态调整高风险房源的保证金比例，实现了从“拍脑袋”到“数据驱动风控”的转变。\n\nngboost 的核心价值在于将机器学习从单纯的“点预测”升级为全面的“概率预测”，让模型不仅能告诉你会发生什么，还能告诉你有多大把握。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fstanfordmlgroup_ngboost_9880fe42.png","stanfordmlgroup","Stanford Machine Learning Group","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fstanfordmlgroup_127c1aa0.png","Our mission is to significantly improve people's lives through our work in AI",null,"mlgroup.stanford.edu","https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup",[22,26,30,34],{"name":23,"color":24,"percentage":25},"Jupyter Notebook","#DA5B0B",85.7,{"name":27,"color":28,"percentage":29},"Python","#3572A5",13.9,{"name":31,"color":32,"percentage":33},"Shell","#89e051",0.3,{"name":35,"color":36,"percentage":37},"Makefile","#427819",0,1857,246,"2026-03-31T10:38:54","Apache-2.0",1,"未说明",{"notes":45,"python":43,"dependencies":46},"该工具基于 Scikit-Learn 构建，可通过 pip 或 conda-forge 安装。README 中未明确指定操作系统、GPU、内存及具体 Python 版本要求，通常意味着它依赖于 Scikit-Learn 的标准运行环境，适用于常规 CPU 环境。",[47],"scikit-learn",[49],"开发框架",[51,52,53,54,6,55],"machine-learning","gradient-boosting","natural-gradients","uncertainty-estimation","python",2,"ready","2026-03-27T02:49:30.150509","2026-04-06T06:46:00.365379",[61,66,71,76,81,86,90],{"id":62,"question_zh":63,"answer_zh":64,"source_url":65},13484,"如何设置随机种子以确保 NGBoost 模型训练结果的可复现性？","仅设置 `np.random.seed` 可能不足以完全复现结果，因为底层库（如 scikit-learn 的决策树）也有自己的随机状态。建议在初始化 `NGBoost` 时显式传递 `random_state` 参数。例如：`NGBoost(..., random_state=2334)`。此外，确保所有相关依赖库的版本一致，并且数据划分（如 `train_test_split`）也设置了相同的 `random_state`。","https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fissues\u002F22",{"id":67,"question_zh":68,"answer_zh":69,"source_url":70},13485,"NGBoost 输出的 'sigma'（标准差）代表的是偶然误差（Aleatoric）、认知误差（Epistemic）还是两者的总和？","NGBoost 直接建模的是条件分布 P(Y|X)，其输出的 'sigma' 主要反映了给定输入 X 下目标变量 Y 的固有随机性，这通常对应于偶然误差（Aleatoric uncertainty）。它并不直接包含由于模型本身不确定性导致的认知误差（Epistemic uncertainty）。如果需要估计认知误差，通常需要结合集成方法（如 Bootstrap 聚合）来评估模型参数的变动范围。","https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fissues\u002F133",{"id":72,"question_zh":73,"answer_zh":74,"source_url":75},13486,"如果遇到 numpy >= 2.0 版本导致的报错（如 ValueError: solve: Input operand 1 has a mismatch...），该如何解决？","该问题已在 NGBoost v0.5.2 版本中修复。请升级您的 NGBoost 包到最新版本：`pip install --upgrade ngboost`。如果暂时无法升级 NGBoost，可以将 numpy 版本回退到 1.26.4 或更低版本：`pip install numpy==1.26.4`。","https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fissues\u002F358",{"id":77,"question_zh":78,"answer_zh":79,"source_url":80},13487,"PyPI 上的 NGBoost 包版本与 GitHub 上的代码标签不一致怎么办？","如果发现 PyPI 上的包版本（如 v0.3.12）与 GitHub 源码不符，通常是因为发布延迟。维护者通常会很快发布修正版本（如随后的 v0.3.13）。解决方法是等待并安装最新的 PyPI 版本，或者直接从 GitHub 安装最新开发版：`pip install git+https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost.git`。","https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fissues\u002F283",{"id":82,"question_zh":83,"answer_zh":84,"source_url":85},13488,"NGBoost 是否有 R 语言的支持计划？","目前 NGBoost 核心库主要是 Python 实现。虽然社区有过关于 R 支持的讨论，但官方尚未提供原生的 R 包。对于需要在 R 中使用类似功能的用户，可以考虑通过 `reticulate` 包在 R 中调用 Python 的 NGBoost 库，或者关注社区是否有人开发了非官方的 R 接口。","https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fissues\u002F14",{"id":87,"question_zh":88,"answer_zh":89,"source_url":70},13489,"如果输入变量（X）本身带有测量误差，如何在 NGBoost 中进行误差传播？","NGBoost 本身不直接处理输入变量的误差传播。如果输入变量 X 具有不确定性（例如服从正态分布），一种通用的方法是使用数据增强（Data Augmentation）技术：从输入变量 X 的概率密度函数（PDF）中抽取多个样本，构建增强的数据集进行训练。这种方法可以将输入的不确定性纳入到模型对 P(Y|X) 的估计过程中，从而间接反映在输出的预测分布中。",{"id":91,"question_zh":92,"answer_zh":93,"source_url":94},13490,"在哪里可以找到 NGBoost 的详细文档和使用示例（Vignettes）？","NGBoost 的用户-facing 方法已添加了详细的 docstrings，您可以直接在 Python 中使用 `help()` 函数查看（例如 `help(NGBoost)`）。此外，项目仓库中通常包含 Jupyter Notebook 格式的 Vignettes 或示例脚本，展示了从基础回归、分类到自定义分布的各种用法。建议查看 GitHub 仓库根目录下的 `examples` 文件夹或官方文档网站（如果有托管）。","https:\u002F\u002Fgithub.com\u002Fstanfordmlgroup\u002Fngboost\u002Fissues\u002F60",[96,101,106,111,116,121,126,131,136,141,146,151,156,161,166,171,175,179,183,187],{"id":97,"version":98,"summary_zh":99,"released_at":100},72242,"v0.5.10","## 版本 0.5.10\r\n\r\n* 添加 `load_ngboost_model` 兼容性加载器，用于加载使用 scikit-learn \u003C 1.3 保存、但在较新版本 scikit-learn 中加载的模型（问题 #389）\r\n* 增加针对旧式树节点 pickle 的定向回归测试，以及针对 `Y_from_censored` 和兼容性加载路径的辅助级别测试","2026-03-24T02:02:35",{"id":102,"version":103,"summary_zh":104,"released_at":105},72243,"v0.5.9","## 版本 0.5.9\n\n* 增加基于 SymPy 的分布工厂支持，允许从符号表达式或 `sympy.stats` 分布定义 LogScore 和 NGBoost 分布\n* 通过该工厂新增内置分布，包括 Beta 分布、Beta-Bernoulli 分布、Beta-二项分布和 Logit-正态分布\n* 添加对 Python 3.14 的支持，并更新 CI 矩阵以测试 Python 3.14","2026-02-25T02:38:48",{"id":107,"version":108,"summary_zh":109,"released_at":110},72244,"v0.5.8","## 版本 0.5.8\r\n\r\n* 修复 numpy2 与自然梯度的兼容性问题","2025-11-21T03:09:05",{"id":112,"version":113,"summary_zh":114,"released_at":115},72245,"v0.5.7","## 版本 0.5.7\n\n* 支持 Python 3.13\n* 更新开发依赖项","2025-10-07T02:28:13",{"id":117,"version":118,"summary_zh":119,"released_at":120},72246,"v0.5.6","## 版本 0.5.6\n\n* 添加对威布尔分布和半正态分布的支持","2025-06-15T00:58:04",{"id":122,"version":123,"summary_zh":124,"released_at":125},72247,"v0.5.5","## 版本 0.5.5\n\n* 升级 sklearn 至 1.6 或更高版本","2025-03-01T19:43:49",{"id":127,"version":128,"summary_zh":129,"released_at":130},72248,"v0.5.4","# 发布说明\n\n## 版本 0.5.4\n\n* 更新为部分拟合，以尊重验证数据\n","2025-02-13T03:24:10",{"id":132,"version":133,"summary_zh":134,"released_at":135},72249,"v0.5.3","## 版本 0.5.3\n\n* 允许输入 NAN\n* 更新 Poetry 依赖\n","2025-02-01T04:39:58",{"id":137,"version":138,"summary_zh":139,"released_at":140},72250,"v0.5.2","## 版本 0.5.2\r\n\r\n* 支持 NumPy 2.0\r\n* 修复了值错误\r\n* 代码风格检查更新","2025-01-23T03:10:29",{"id":142,"version":143,"summary_zh":144,"released_at":145},72251,"v0.5.1","## 版本 0.5.1\n\n* 增加对 NormalFixedMean 分布的支持\n* 更新 Makefile，以便更轻松地发布","2024-02-21T02:20:47",{"id":147,"version":148,"summary_zh":149,"released_at":150},72252,"v0.5.0","Version 0.5.0\r\n\r\nDrops support for python 3.7 and 3.8\r\nNow supports Python 3.11 and 3.12\r\nFixed issue with np.bool\r\nOptimized memory usage in pred-dist\r\nRemoved declared pandas dependency\r\nSignificant improvements to run times on tests during development\r\nMinor enhancements to github actions","2024-02-14T01:12:31",{"id":152,"version":153,"summary_zh":154,"released_at":155},72253,"v0.4.2","Fix deprecated numpy type alias. This was causing a warning with NumPy >=1.20 and an error with NumPy >=1.24\r\nRemove pandas as a declared dependency","2023-11-01T03:50:41",{"id":157,"version":158,"summary_zh":159,"released_at":160},72254,"v0.4.1","Version 0.4.1\r\nAdded `partial_fit` method for incremental learning\r\n\r\n NGBoost now includes a new `partial_fit` method that allows for incremental learning. This method appends new base models to the existing ones, which can be useful when new data becomes available over time or when the data is too large to fit in memory all at once.\r\n\r\n The `partial_fit` method takes similar parameters to the `fit` method, including predictors `X`, outcomes `Y`, and validation sets `X_val` and `Y_val`. It also supports custom weights for the training and validation sets, as well as early stopping and custom loss monitoring.\r\n\r\n Please note that the `partial_fit` method is not yet fully tested and may not work as expected in all cases. Use it with caution and thoroughly test its behavior in your specific use case before relying on it in production.","2023-03-31T19:42:16",{"id":162,"version":163,"summary_zh":164,"released_at":165},72255,"v0.4.0","For release notes see RELEASE_NOTES.md","2023-03-14T21:21:55",{"id":167,"version":168,"summary_zh":169,"released_at":170},72256,"v0.3.13","Release Notes 0.3.13\r\n\r\n- Update to drop support for Python 3.6 and add support for 3.9 and 3.10\r\n- Update black formatter to latest\r\n- Fix bug with previous code mismatch\r\n","2022-10-13T15:00:35",{"id":172,"version":173,"summary_zh":18,"released_at":174},72257,"v0.3.12","2021-07-30T15:13:23",{"id":176,"version":177,"summary_zh":18,"released_at":178},72258,"v0.3.11","2021-06-03T16:17:00",{"id":180,"version":181,"summary_zh":18,"released_at":182},72259,"v0.3.10","2021-03-25T14:39:56",{"id":184,"version":185,"summary_zh":18,"released_at":186},72260,"v0.3.9","2021-02-25T17:30:20",{"id":188,"version":189,"summary_zh":18,"released_at":190},72261,"v0.3.8","2021-02-23T02:32:04",[192,203,212,220,228,241],{"id":193,"name":194,"github_repo":195,"description_zh":196,"stars":197,"difficulty_score":198,"last_commit_at":199,"category_tags":200,"status":57},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",[49,201,202],"图像","Agent",{"id":204,"name":205,"github_repo":206,"description_zh":207,"stars":208,"difficulty_score":56,"last_commit_at":209,"category_tags":210,"status":57},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,"2026-04-05T11:33:21",[49,202,211],"语言模型",{"id":213,"name":214,"github_repo":215,"description_zh":216,"stars":217,"difficulty_score":56,"last_commit_at":218,"category_tags":219,"status":57},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",[49,201,202],{"id":221,"name":222,"github_repo":223,"description_zh":224,"stars":225,"difficulty_score":56,"last_commit_at":226,"category_tags":227,"status":57},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",[49,211],{"id":229,"name":230,"github_repo":231,"description_zh":232,"stars":233,"difficulty_score":56,"last_commit_at":234,"category_tags":235,"status":57},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",[201,236,237,238,202,239,211,49,240],"数据工具","视频","插件","其他","音频",{"id":242,"name":243,"github_repo":244,"description_zh":245,"stars":246,"difficulty_score":198,"last_commit_at":247,"category_tags":248,"status":57},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",[202,201,49,211,239]]