[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-itdxer--neupy":3,"tool-itdxer--neupy":61},[4,18,26,36,44,53],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":10,"last_commit_at":24,"category_tags":25,"status":17},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,"2026-04-05T11:01:52",[14,15,13],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"status":17},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",150037,2,"2026-04-10T23:33:47",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"status":17},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",108322,"2026-04-10T11:39:34",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":32,"last_commit_at":59,"category_tags":60,"status":17},4721,"markitdown","microsoft\u002Fmarkitdown","MarkItDown 是一款由微软 AutoGen 团队打造的轻量级 Python 工具，专为将各类文件高效转换为 Markdown 格式而设计。它支持 PDF、Word、Excel、PPT、图片（含 OCR）、音频（含语音转录）、HTML 乃至 YouTube 链接等多种格式的解析，能够精准提取文档中的标题、列表、表格和链接等关键结构信息。\n\n在人工智能应用日益普及的今天，大语言模型（LLM）虽擅长处理文本，却难以直接读取复杂的二进制办公文档。MarkItDown 恰好解决了这一痛点，它将非结构化或半结构化的文件转化为模型“原生理解”且 Token 效率极高的 Markdown 格式，成为连接本地文件与 AI 分析 pipeline 的理想桥梁。此外，它还提供了 MCP（模型上下文协议）服务器，可无缝集成到 Claude Desktop 等 LLM 应用中。\n\n这款工具特别适合开发者、数据科学家及 AI 研究人员使用，尤其是那些需要构建文档检索增强生成（RAG）系统、进行批量文本分析或希望让 AI 助手直接“阅读”本地文件的用户。虽然生成的内容也具备一定可读性，但其核心优势在于为机器",93400,"2026-04-06T19:52:38",[52,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":66,"readme_en":67,"readme_zh":68,"quickstart_zh":69,"use_case_zh":70,"hero_image_url":71,"owner_login":72,"owner_name":73,"owner_avatar_url":74,"owner_bio":75,"owner_company":75,"owner_location":75,"owner_email":75,"owner_twitter":75,"owner_website":76,"owner_url":77,"languages":78,"stars":98,"forks":99,"last_commit_at":100,"license":101,"difficulty_score":32,"env_os":102,"env_gpu":103,"env_ram":104,"env_deps":105,"category_tags":109,"github_topics":110,"view_count":32,"oss_zip_url":75,"oss_zip_packed_at":75,"status":17,"created_at":123,"updated_at":124,"faqs":125,"releases":154},6575,"itdxer\u002Fneupy","neupy","NeuPy is a Tensorflow based python library for prototyping and building neural networks","NeuPy 是一个基于 TensorFlow 构建的 Python 库，旨在帮助开发者快速原型设计和搭建神经网络。它主要解决了在深度学习模型开发初期，需要灵活、简洁地验证算法想法而不必陷入复杂底层代码的问题。通过封装 TensorFlow 的计算能力，NeuPy 让神经网络的构建过程更加直观和高效。\n\n这款工具特别适合有一定编程基础的开发者、数据科学家以及研究人员使用，尤其是那些希望快速尝试不同神经网络架构或学习经典算法原理的用户。NeuPy 的独特亮点在于其不仅支持常见的深度学习模型，还内置了多种经典的无监督学习算法，如“生长神经气体”（Growing Neural Gas）和“自组织映射”（SOFM）。这些算法能够自动学习数据的拓扑结构，甚至能用于生成独特的艺术图案，展现了其在探索性数据分析和创意编码方面的潜力。\n\n需要注意的是，NeuPy 目前已不再积极维护，因此在将其应用于大型生产项目前，建议充分评估其长期兼容性。但对于教学演示、算法研究或小型实验项目而言，NeuPy 依然是一个轻量且富有启发性的选择。","|Travis|_ |Coverage|_ |Dependency Status|_ |License|_\n\n.. |Travis| image:: https:\u002F\u002Fapi.travis-ci.org\u002Fitdxer\u002Fneupy.png?branch=master\n.. _Travis: https:\u002F\u002Ftravis-ci.org\u002Fitdxer\u002Fneupy\n\n.. |Dependency Status| image:: https:\u002F\u002Fdependencyci.com\u002Fgithub\u002Fitdxer\u002Fneupy\u002Fbadge\n.. _Dependency Status: https:\u002F\u002Fdependencyci.com\u002Fgithub\u002Fitdxer\u002Fneupy\n\n.. |License| image:: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-MIT-blue.svg\n.. _License: https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002FLICENSE\n\n.. |Coverage| image:: https:\u002F\u002Fcodecov.io\u002Fgh\u002Fitdxer\u002Fneupy\u002Fbranch\u002Fmaster\u002Fgraph\u002Fbadge.svg\n.. _Coverage: https:\u002F\u002Fcodecov.io\u002Fgh\u002Fitdxer\u002Fneupy\n\n\n \\:warning: **The library is no longer being actively maintained, please consider this before using it for your project** :warning:\n\n\n.. raw:: html\n\n    \u003Cdiv align=\"center\">\n        \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F17\u002Fsofm_text_style.html\">\n        \u003Cimg width=\"80%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_56dd8964ef64.png\">\n        \u003C\u002Fa>\n    \u003C\u002Fdiv>\n\n\nNeuPy v0.8.2\n============\n\nNeuPy is a python library for prototyping and building neural networks. NeuPy uses Tensorflow as a computational backend for deep learning models.\n\nInstallation\n------------\n\n.. code-block:: bash\n\n    $ pip install neupy\n\nUser Guide\n----------\n\n* `Install NeuPy \u003Chttp:\u002F\u002Fneupy.com\u002Fpages\u002Finstallation.html>`_\n* Check the `tutorials \u003Chttp:\u002F\u002Fneupy.com\u002Fdocs\u002Ftutorials.html>`_\n* Learn more about NeuPy in the `documentation \u003Chttp:\u002F\u002Fneupy.com\u002Fpages\u002Fdocumentation.html>`_\n* Explore lots of different `neural network algorithms \u003Chttp:\u002F\u002Fneupy.com\u002Fpages\u002Fcheatsheet.html>`_.\n* Read `articles \u003Chttp:\u002F\u002Fneupy.com\u002Farchive.html>`_ and learn more about Neural Networks.\n* `Open Issues \u003Chttps:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fissues>`_ and ask questions.\n\nArticles and Notebooks\n----------------------\n\n.. raw:: html\n\n    \u003Ctable border=\"0\">\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2018\u002F03\u002F26\u002Fmaking_art_with_growing_neural_gas.html#id1\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_d4e8785e72d5.gif\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2018\u002F03\u002F26\u002Fmaking_art_with_growing_neural_gas.html#id1\">Growing Neural Gas\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>Growing Neural Gas is an algorithm that learns topological structure of the data.\u003C\u002Fp>\n                \u003Cp>Code that generates animation can be found in \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002Fgrowing-neural-gas\u002FGrowing%20Neural%20Gas%20animated.ipynb\">this ipython notebook\u003C\u002Fa>\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2018\u002F03\u002F26\u002Fmaking_art_with_growing_neural_gas.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_d564021142d7.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2018\u002F03\u002F26\u002Fmaking_art_with_growing_neural_gas.html\">Making Art with Growing Neural Gas\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>Growing Neural Gas is another example of the algorithm that follows simple set of rules that on a large scale can generate complex patterns.\u003C\u002Fp>\n                \u003Cp>Image on the left is a great example of the art style that can be generated with simple set fo rules.\u003C\u002Fp>\n                \u003Cp>The main notebook that generates image can be found \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002Fgrowing-neural-gas\u002FMaking%20Art%20with%20Growing%20Neural%20Gas.ipynb\">here\u003C\u002Fa>\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F09\u002Fsofm_applications.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_cbdf264d918c.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F09\u002Fsofm_applications.html\">Self-Organizing Maps and Applications\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>\n                    Self-Organazing Maps (SOM or SOFM) is a very simple and powerful algorithm that has a wide variety of applications. This articles covers some of them, including:\n\n                    \u003Cul>\n                        \u003Cli>Visualizing Convolutional Neural Networks\u003C\u002Fli>\n                        \u003Cli>Data topology learning\u003C\u002Fli>\n                        \u003Cli>High-dimensional data visualization\u003C\u002Fli>\n                        \u003Cli>Clustering\u003C\u002Fli>\n                    \u003C\u002Ful>\n                \u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002FVisualizing%20CNN%20based%20on%20Pre-trained%20VGG19.ipynb\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_fbf204a97912.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002FVisualizing%20CNN%20based%20on%20Pre-trained%20VGG19.ipynb\">Visualizing CNN based on Pre-trained VGG19\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>This notebook shows how you can easely explore reasons behind convolutional network predictions and understand what type of features has been learned in different layers of the network.\u003C\u002Fp>\n                \u003Cp>In addition, this notebook shows how to use neural network architectures in NeuPy, like VGG19, with pre-trained parameters.\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2015\u002F07\u002F04\u002Fvisualize_backpropagation_algorithms.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_21083aa64d53.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2015\u002F07\u002F04\u002Fvisualize_backpropagation_algorithms.html\">Visualize Algorithms based on the Backpropagation\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>Image on the left shows comparison between paths that different algorithm take along the descent path. It's interesting to see how much information about the algorithm can be extracted from simple trajectory paths. All of this covered and explained in the article.\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2016\u002F12\u002F17\u002Fhyperparameter_optimization_for_neural_networks.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_0dcb9017c685.png\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_098bda7ee880.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2016\u002F12\u002F17\u002Fhyperparameter_optimization_for_neural_networks.html\">Hyperparameter optimization for Neural Networks\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>\n                    This article covers different approaches for hyperparameter optimization.\n                    \u003Cul>\n                    \u003Cli>Grid Search\u003C\u002Fli>\n                    \u003Cli>Random Search\u003C\u002Fli>\n                    \u003Cli>Hand-tuning\u003C\u002Fli>\n                    \u003Cli>Gaussian Process with Expected Improvement\u003C\u002Fli>\n                    \u003Cli>Tree-structured Parzen Estimators (TPE)\u003C\u002Fli>\n                    \u003C\u002Ful>\n                \u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F13\u002Fsofm_art.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_d5a85ec47266.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F13\u002Fsofm_art.html\">The Art of SOFM\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>In this article, I just want to show how beautiful sometimes can be a neural network. I think, it’s quite rare that algorithm can not only extract knowledge from the data, but also produce something beautiful using exactly the same set of training rules without any modifications.\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2015\u002F09\u002F20\u002Fdiscrete_hopfield_network.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_c5cf8e01ad3c.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2015\u002F09\u002F20\u002Fdiscrete_hopfield_network.html\">Discrete Hopfield Network\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>Article with extensive theoretical background about Discrete Hopfield Network. It also has example that show advantages and limitations of the algorithm.\u003C\u002Fp>\n                \u003Cp>Image on the left is a visulatization of the information stored in the network. This picture not only visualizes network's memory, ot shows everything network knows about the world.\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F17\u002Fsofm_text_style.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_cbca7f5d5152.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F17\u002Fsofm_text_style.html\">Create unique text-style with SOFM\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>This article describes step-by-step solution that allow to generate unique styles with arbitrary text.\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002FPlaying%20with%20MLP%20visualizations.ipynb\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_dd171f064128.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002FPlaying%20with%20MLP%20visualizations.ipynb\">Playing with MLP visualizations\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>This notebook shows interesting ways to look inside of your MLP network.\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Ftree\u002Fmaster\u002Fexamples\u002Freinforcement_learning\u002Fvin\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_8af2b23c2e0c.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Ftree\u002Fmaster\u002Fexamples\u002Freinforcement_learning\u002Fvin\">Exploring world with Value Iteration Network (VIN)\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>One of the basic applications of the Value Iteration Network that learns how to find an optimal path between two points in the environment with obstacles.\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Ftree\u002Fmaster\u002Fexamples\u002Fboltzmann_machine\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_fc4d61f9ede7.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Ftree\u002Fmaster\u002Fexamples\u002Fboltzmann_machine\">Features learned by Restricted Boltzmann Machine (RBM)\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>Set of examples that use and explore knowledge extracted by Restricted Boltzmann Machine\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n    \u003C\u002Ftable>\n","|Travis|_ |覆盖率|_ |依赖状态|_ |许可证|_\n\n.. |Travis| image:: https:\u002F\u002Fapi.travis-ci.org\u002Fitdxer\u002Fneupy.png?branch=master\n.. _Travis: https:\u002F\u002Ftravis-ci.org\u002Fitdxer\u002Fneupy\n\n.. |依赖状态| image:: https:\u002F\u002Fdependencyci.com\u002Fgithub\u002Fitdxer\u002Fneupy\u002Fbadge\n.. _依赖状态: https:\u002F\u002Fdependencyci.com\u002Fgithub\u002Fitdxer\u002Fneupy\n\n.. |许可证| image:: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-MIT-blue.svg\n.. _许可证: https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002FLICENSE\n\n.. |覆盖率| image:: https:\u002F\u002Fcodecov.io\u002Fgh\u002Fitdxer\u002Fneupy\u002Fbranch\u002Fmaster\u002Fgraph\u002Fbadge.svg\n.. _覆盖率: https:\u002F\u002Fcodecov.io\u002Fgh\u002Fitdxer\u002Fneupy\n\n\n \\:warning: **该库已不再积极维护，请在将其用于您的项目之前慎重考虑** :warning:\n\n\n.. raw:: html\n\n    \u003Cdiv align=\"center\">\n        \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F17\u002Fsofm_text_style.html\">\n        \u003Cimg width=\"80%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_56dd8964ef64.png\">\n        \u003C\u002Fa>\n    \u003C\u002Fdiv>\n\n\nNeuPy v0.8.2\n============\n\nNeuPy 是一个用于神经网络原型设计和构建的 Python 库。NeuPy 使用 TensorFlow 作为深度学习模型的计算后端。\n\n安装\n----\n\n.. code-block:: bash\n\n    $ pip install neupy\n\n用户指南\n--------\n\n* `安装 NeuPy \u003Chttp:\u002F\u002Fneupy.com\u002Fpages\u002Finstallation.html>`_\n* 查看 `教程 \u003Chttp:\u002F\u002Fneupy.com\u002Fdocs\u002Ftutorials.html>`_\n* 在 `文档 \u003Chttp:\u002F\u002Fneupy.com\u002Fpages\u002Fdocumentation.html>`_ 中了解更多关于 NeuPy 的信息\n* 探索各种不同的 `神经网络算法 \u003Chttp:\u002F\u002Fneupy.com\u002Fpages\u002Fcheatsheet.html>`_。\n* 阅读 `文章 \u003Chttp:\u002F\u002Fneupy.com\u002Farchive.html>`_，进一步了解神经网络。\n* 浏览 `问题列表 \u003Chttps:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fissues>`_ 并提出问题。\n\n文章与笔记本\n--------------\n\n.. raw:: html\n\n    \u003Ctable border=\"0\">\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2018\u002F03\u002F26\u002Fmaking_art_with_growing_neural_gas.html#id1\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_d4e8785e72d5.gif\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2018\u002F03\u002F26\u002Fmaking_art_with_growing_neural_gas.html#id1\">生长型神经气\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>生长型神经气是一种能够学习数据拓扑结构的算法。\u003C\u002Fp>\n                \u003Cp>生成动画的代码可以在 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002Fgrowing-neural-gas\u002FGrowing%20Neural%20Gas%20animated.ipynb\">这个 IPython 笔记本\u003C\u002Fa>中找到。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2018\u002F03\u002F26\u002Fmaking_art_with_growing_neural_gas.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_d564021142d7.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2018\u002F03\u002F26\u002Fmaking_art_with_growing_neural_gas.html\">用生长型神经气创作艺术\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>生长型神经气是另一个遵循简单规则集的算法示例，在大规模下可以生成复杂的模式。\u003C\u002Fp>\n                \u003Cp>左侧的图像就是一个由简单规则集生成的艺术风格的绝佳例子。\u003C\u002Fp>\n                \u003Cp>生成该图像的主要笔记本可以在 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002Fgrowing-neural-gas\u002FMaking%20Art%20with%20Growing%20Neural%20Gas.ipynb\">这里\u003C\u002Fa>找到。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F09\u002Fsofm_applications.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_cbdf264d918c.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F09\u002Fsofm_applications.html\">自组织映射及其应用\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>\n                    自组织映射（SOM 或 SOFM）是一种非常简单而强大的算法，具有广泛的应用。本文介绍了一些应用，包括：\n\n\u003Cul>\n                        \u003Cli>卷积神经网络的可视化\u003C\u002Fli>\n                        \u003Cli>数据拓扑结构学习\u003C\u002Fli>\n                        \u003Cli>高维数据可视化\u003C\u002Fli>\n                        \u003Cli>聚类\u003C\u002Fli>\n                    \u003C\u002Ful>\n                \u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002FVisualizing%20CNN%20based%20on%20Pre-trained%20VGG19.ipynb\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_fbf204a97912.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002FVisualizing%20CNN%20based%20on%20Pre-trained%20VGG19.ipynb\">基于预训练VGG19的CNN可视化\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>本笔记本展示了如何轻松探索卷积网络预测背后的原因，并理解网络不同层中学习到了哪些类型的特征。\u003C\u002Fp>\n                \u003Cp>此外，本笔记本还演示了如何在NeuPy中使用具有预训练参数的神经网络架构，例如VGG19。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2015\u002F07\u002F04\u002Fvisualize_backpropagation_algorithms.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_21083aa64d53.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2015\u002F07\u002F04\u002Fvisualize_backpropagation_algorithms.html\">基于反向传播算法的可视化\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>左侧图片展示了不同算法在下降路径上所走的轨迹对比。有趣的是，仅从简单的轨迹路径中就能提取出关于算法的大量信息。文章对这些内容进行了全面介绍和解释。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2016\u002F12\u002F17\u002Fhyperparameter_optimization_for_neural_networks.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_0dcb9017c685.png\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_098bda7ee880.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2016\u002F12\u002F17\u002Fhyperparameter_optimization_for_neural_networks.html\">神经网络的超参数优化\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>\n                    本文介绍了多种超参数优化方法。\n                    \u003Cul>\n                    \u003Cli>网格搜索\u003C\u002Fli>\n                    \u003Cli>随机搜索\u003C\u002Fli>\n                    \u003Cli>手动调参\u003C\u002Fli>\n                    \u003Cli>基于期望改进的高斯过程\u003C\u002Fli>\n                    \u003Cli>树结构Parzen估计器（TPE）\u003C\u002Fli>\n                    \u003C\u002Ful>\n                \u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F13\u002Fsofm_art.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_d5a85ec47266.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F13\u002Fsofm_art.html\">SOFM的艺术\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>在这篇文章中，我想展示神经网络有时可以多么美丽。我认为，很少有算法不仅能从数据中提取知识，还能在不作任何修改的情况下，利用相同的训练规则创造出美丽的作品。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2015\u002F09\u002F20\u002Fdiscrete_hopfield_network.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_c5cf8e01ad3c.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2015\u002F09\u002F20\u002Fdiscrete_hopfield_network.html\">离散霍普菲尔德网络\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>本文提供了关于离散霍普菲尔德网络的丰富理论背景，并包含一个示例，展示了该算法的优势和局限性。\u003C\u002Fp>\n                \u003Cp>左侧图片是网络中存储信息的可视化。这张图不仅展示了网络的记忆，也呈现了网络对世界的全部认知。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F17\u002Fsofm_text_style.html\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_cbca7f5d5152.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"http:\u002F\u002Fneupy.com\u002F2017\u002F12\u002F17\u002Fsofm_text_style.html\">用SOFM创建独特的文本风格\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>本文描述了一种逐步解决方案，可用于生成任意文本的独特风格。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002FPlaying%20with%20MLP%20visualizations.ipynb\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_dd171f064128.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fblob\u002Fmaster\u002Fnotebooks\u002FPlaying%20with%20MLP%20visualizations.ipynb\">玩转MLP可视化\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>本笔记本展示了深入观察MLP网络内部的有趣方法。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Ftree\u002Fmaster\u002Fexamples\u002Freinforcement_learning\u002Fvin\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_8af2b23c2e0c.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Ftree\u002Fmaster\u002Fexamples\u002Freinforcement_learning\u002Fvin\">用价值迭代网络（VIN）探索世界\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>这是价值迭代网络的一个基本应用，它能够学习在存在障碍物的环境中找到两点之间的最优路径。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n        \u003Ctr>\n            \u003Ctd border=\"0\" width=\"30%\">\n                \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Ftree\u002Fmaster\u002Fexamples\u002Fboltzmann_machine\">\n                \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_readme_fc4d61f9ede7.png\">\n                \u003C\u002Fa>\n            \u003C\u002Ftd>\n            \u003Ctd border=\"0\" valign=\"top\">\n                \u003Ch3>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Ftree\u002Fmaster\u002Fexamples\u002Fboltzmann_machine\">受限玻尔兹曼机（RBM）学到的特征\u003C\u002Fa>\u003C\u002Fh3>\n                \u003Cp>这是一组使用并探索受限玻尔兹曼机所提取知识的示例。\u003C\u002Fp>\n            \u003C\u002Ftd>\n        \u003C\u002Ftr>\n    \u003C\u002Ftable>","# NeuPy 快速上手指南\n\n> **⚠️ 重要提示**：该库目前已不再积极维护。在将其用于生产项目前，请充分考虑此风险。建议仅用于学习、原型验证或研究特定经典算法（如 SOM、GNG 等）。\n\nNeuPy 是一个基于 Python 的神经网络原型构建库，底层使用 TensorFlow 作为计算后端，特别适合快速实现和实验各种神经网络算法。\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 3.6 - 3.8（鉴于库的维护状态，过高版本的 Python 可能存在兼容性问题）\n*   **前置依赖**：\n    *   `pip` (Python 包管理工具)\n    *   `TensorFlow` (NeuPy 将自动处理或依赖其进行计算)\n\n## 2. 安装步骤\n\n您可以使用 pip 直接安装 NeuPy。国内开发者建议使用清华或阿里镜像源以加速下载。\n\n**标准安装命令：**\n```bash\npip install neupy\n```\n\n**使用国内镜像源加速安装（推荐）：**\n```bash\npip install neupy -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n## 3. 基本使用\n\nNeupy 的设计初衷是简化神经网络的构建过程。以下是一个最简单的示例，展示如何创建一个多层感知机（MLP）并进行训练。\n\n### 简单示例：构建并训练一个 MLP\n\n```python\nfrom neupy import algorithms\n\n# 定义训练数据 (异或问题 XOR)\ninput_data = [[0, 0], [0, 1], [1, 0], [1, 1]]\ntarget_data = [[0], [1], [1], [0]]\n\n# 初始化网络\n# 架构：2 个输入神经元 -> 5 个隐藏层神经元 (Sigmoid 激活) -> 1 个输出神经元\nnetwork = algorithms.BackPropagation(\n    (2, 5, 1),\n    verbose=False,\n    step=0.1,\n    shuffle_data=True\n)\n\n# 训练网络\nnetwork.train(input_data, target_data, epochs=1000)\n\n# 预测结果\nprediction = network.predict(input_data)\nprint(\"预测结果:\")\nprint(prediction)\n```\n\n### 下一步学习\n由于本库功能丰富但已停止更新，建议参考以下资源深入理解特定算法的实现细节：\n*   **官方教程**：访问 [NeuPy Tutorials](http:\u002F\u002Fneupy.com\u002Fdocs\u002Ftutorials.html) 查看更详细的用例。\n*   **算法速查表**：查阅 [Cheatsheet](http:\u002F\u002Fneupy.com\u002Fpages\u002Fcheatsheet.html) 了解支持的神经网络算法列表（包括 SOM, GNG, RBM 等）。\n*   **示例代码**：探索 GitHub 仓库中的 `notebooks` 和 `examples` 目录，那里包含了大量关于可视化、艺术生成及强化学习的完整代码示例。","某数据科学团队正在探索利用自组织映射（SOFM）和生长神经气体（GNG）算法，对高维客户行为数据进行无监督聚类与可视化分析。\n\n### 没有 neupy 时\n- **底层代码繁琐**：开发者需直接使用 TensorFlow 原生 API 从零构建复杂的无监督网络拓扑，大量时间耗费在矩阵运算和梯度更新逻辑上。\n- **算法复现困难**：像生长神经气体这类非标准深度学习算法，缺乏现成模块，查阅论文并手动复现规则极易出错且耗时漫长。\n- **原型迭代缓慢**：每次调整网络结构或超参数都需要重写大量样板代码，导致验证不同聚类效果的开发周期长达数天。\n- **可视化门槛高**：将高维拓扑结构转化为直观的二维动态图需要额外编写复杂的绘图脚本，难以快速向业务方展示数据分布特征。\n\n### 使用 neupy 后\n- **高级封装提效**：neupy 提供了基于 TensorFlow 的高级抽象，仅需几行代码即可实例化 SOFM 或 GNG 模型，自动处理底层计算细节。\n- **内置丰富算法**：直接调用库中预置的生长神经气体等稀有算法，无需手动复现论文公式，确保逻辑准确且开箱即用。\n- **快速原型验证**：通过简洁的 API 灵活调整网络节点数和邻域函数，团队能在几小时内完成多种拓扑结构的对比实验。\n- **集成示例支持**：借助官方提供的 Jupyter Notebook 案例，轻松生成类似“艺术风格化”的动态拓扑演化图，直观呈现数据内在结构。\n\nneupy 的核心价值在于将复杂的无监督神经网络算法封装为易用的原型工具，让研究者能专注于数据洞察而非底层实现。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fitdxer_neupy_5170cc58.png","itdxer","Yurii","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fitdxer_865ceb58.png",null,"http:\u002F\u002Fblog.itdxer.com\u002F","https:\u002F\u002Fgithub.com\u002Fitdxer",[79,83,87,91,94],{"name":80,"color":81,"percentage":82},"Python","#3572A5",50.7,{"name":84,"color":85,"percentage":86},"Jupyter Notebook","#DA5B0B",49.2,{"name":88,"color":89,"percentage":90},"HTML","#e34c26",0.1,{"name":92,"color":93,"percentage":90},"JavaScript","#f1e05a",{"name":95,"color":96,"percentage":97},"Shell","#89e051",0,735,160,"2026-04-09T13:54:21","MIT","","未说明（底层依赖 TensorFlow，具体 GPU 需求取决于安装的 TensorFlow 版本）","未说明",{"notes":106,"python":104,"dependencies":107},"该库已不再积极维护，使用前请慎重考虑。NeuPy 使用 TensorFlow 作为深度学习模型的计算后端。安装命令为 'pip install neupy'。",[108],"tensorflow",[14],[111,112,113,114,115,64,108,116,117,118,119,120,121,122],"deep-learning","deep-neural-networks","deeplearning","neural-network","artificial-neural-networks","som","sofm","hopfield","rbm","boltzmann-machine","pnn","lvq","2026-03-27T02:49:30.150509","2026-04-11T17:39:38.757108",[126,131,136,141,146,150],{"id":127,"question_zh":128,"answer_zh":129,"source_url":130},29680,"为什么在 GPU 上使用拟牛顿法（如 BFGS）训练时会出现 NaN 错误？","这是一个已知问题，已在 release\u002Fv0.3.1 分支中修复。你可以通过以下命令安装该修复版本：\npip install git+https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy.git@release\u002Fv0.3.1\n安装后，GPU 上的拟牛顿法算法（BFGS 和 SR1）应能正常工作，不再返回 NaN。","https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fissues\u002F108",{"id":132,"question_zh":133,"answer_zh":134,"source_url":135},29681,"如何确保 NeuPy 神经网络训练结果的可重复性？","结果不可复现通常与随机种子设置或参数空间定义方式有关。建议：\n1. 使用 `environment.reproducible()` 设置全局随机种子。\n2. 在进行超参数优化时，避免直接使用 `np.random`，而应使用 `hp.uniform` 等函数来定义参数空间，以确保分布的正确性和效率。\n3. 确保升级到最新版本（如 v0.5.2），其中修复了导致不可复现的底层逻辑问题。","https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fissues\u002F173",{"id":137,"question_zh":138,"answer_zh":139,"source_url":140},29682,"如何在 NeuPy 中实现自定义激活函数（例如将浮点数转换为定点数）而不破坏梯度传播？","直接使用取整操作（如 `T.floor` 或 `int`）会导致梯度变为零，从而使网络无法学习（误差保持不变）。\n解决方案是采用“直通估计器”（Straight-Through Estimator）的思想：在前向传播时使用量化后的权重（定点数），但在反向传播计算梯度时，假装使用的是原始浮点权重。\n伪代码逻辑如下：\n1. 生成随机权重 weight。\n2. 创建量化权重 quantized_weight = float_limit(weight)。\n3. 前向传播使用 quantized_weight 计算输出。\n4. 反向传播时，将梯度直接传递给 weight，忽略量化操作的导数（即视其导数为 1）。","https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fissues\u002F209",{"id":142,"question_zh":143,"answer_zh":144,"source_url":145},29683,"当数据集存在严重偏差（如大量值为 0）导致模型训练效果不佳时，有什么建议？","如果数据集中存在严重的类别不平衡或数值偏差（例如大量偏向 0），单纯更换激活函数（如从 Sigmoid 换到 ReLU）可能无法根本解决问题，尽管 ReLU 有助于缓解梯度消失。\n建议采取以下步骤：\n1. 检查网络原始输出（未舍入前）与目标值之间的残差。\n2. 分析残差分布，如果残差不是均值为 0 的正态分布，可以尝试寻找更好的方法来转换网络的原始输出，而不是仅仅依赖激活函数。","https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fissues\u002F92",{"id":147,"question_zh":148,"answer_zh":149,"source_url":145},29684,"ReLU 激活函数相比 Sigmoid 有什么优势？它能解决所有训练慢的问题吗？","ReLU 的主要优势在于它不太容易导致梯度消失（即梯度值变得非常接近 0），而梯度消失会使训练变慢。然而，ReLU 并不是万能的，如果你的问题是由数据本身的偏差或其他架构问题引起的，仅切换到 ReLU 可能无法修复该问题。",{"id":151,"question_zh":152,"answer_zh":153,"source_url":135},29685,"在使用 NeuPy 进行超参数搜索时，如何正确定义参数空间？","不要直接使用 `np.random` 来生成离散分布的参数，这会导致搜索空间受限且效率低下。应该使用专门的超参数优化库提供的函数（如 `hp.uniform`）来定义参数空间。这样可以确保参数分布连续且适合优化算法（如 TPE）高效工作。",[155,160,165,170,175,180,185,190,195,200,205,210,215,220,225,230,235,240,245,250],{"id":156,"version":157,"summary_zh":158,"released_at":159},206248,"v0.8.2","主要变更：\n* 改进了层\u002F网络复制功能\n* 添加了 `repeat` 函数\n* 修复了当输入形状未知时卷积层的输出形状问题\n* 修正了拼写错误","2019-04-04T19:44:59",{"id":161,"version":162,"summary_zh":163,"released_at":164},206249,"v0.8.1","主要变更：\n* 添加了对 TensorFlow 1.13 的支持\n* 添加了 DropBlock 层\n* 为所有层添加了 `define` 方法\n","2019-03-30T20:04:11",{"id":166,"version":167,"summary_zh":168,"released_at":169},206250,"v0.8.0","主要变更：\n- 图层图和图层采用全新设计\n- 将内联运算符 `>`（仍支持）改为 `>>`，并将列表用于并行连接的 `|` 运算符\n- 新增两种阶梯式衰减算法，分别为 `exponential_decay` 和 `polynomial_decay`\n- 新增正则化器\n- 引入信号功能，可通过类进行扩展（`signals` 属性）\n- 使用 TensorFlow 优化器替代自实现函数\n- 新增 `GroupNorm` 层\n- 改变了 `predict` 方法的行为\n- 为网络添加了 `show` 方法\n- 为优化器新增 `plot_errors`，提供更优、更精确的可视化效果\n\n小幅变更：\n- 优化器的 `error` 参数更名为 `loss`\n- 更改了日志中的摘要格式\n- 移除了 `training_errors` 和 `validation_errors`，并以新的 `errors` 属性取而代之\n- 将 `prediction_error` 方法更名为 `score`\n\n其他：\n- 大规模代码重构\n- 改进异常处理机制\n\n已移除：\n- 插件被移除\n- 网络配置中不再支持整数元组\n- 移除了 RBFKmeans\n- 移除了 `train_end_signal` 和 `epoch_end_signal`\n- 移除了 `plots.error_plot` 函数\n- 移除了 `plots.network_structure` 函数","2019-01-28T19:47:37",{"id":171,"version":172,"summary_zh":173,"released_at":174},206251,"v0.7.3","* 修复了从 pickle 文件加载 GNG 时出现的问题。相关问题：https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fissues\u002F229\n* 改进了 GNG 图元的表示方式\n* 在 GNG 图的方法中添加了额外的异常，以便在执行意外操作时提供更具信息量的错误提示","2019-01-17T09:26:16",{"id":176,"version":177,"summary_zh":178,"released_at":179},206252,"v0.7.2","变更：\n- 为 ResNet-50 架构添加了缺失的膨胀率\n- 修复了权重初始化器的扇形计算问题\n- 在预测方法中添加了变量初始化器\n- 将默认初始化器由 XavierNormal 更改为 HeNormal\n","2018-12-13T16:40:43",{"id":181,"version":182,"summary_zh":183,"released_at":184},206253,"v0.7.1","变更：\n- 通过使用 TensorFlow 的张量而非 NumPy 数组进行惰性参数初始化，加快网络初始化速度。\n- 全局池化层现在接受两个字符串参数，分别指向不同的 TensorFlow 函数。\n- 修复了在输入形状未知时使用重塑层的问题。\n- 修复了针对空间输入的交叉熵损失函数。\n- 移除了训练过程中的输入阻塞。\n- 将梯度下降法和小批量梯度下降法合并为一个类：GradientDescent。","2018-12-10T15:01:21",{"id":186,"version":187,"summary_zh":188,"released_at":189},206254,"v0.7.0","变更：\n- 后端迁移至 TensorFlow\n- 权重的 Pickle 存储已被 HDF5 替代\n- 卷积滤波器的维度顺序已更改（假设通道位于最后一个维度）\n- `compile` 方法已被移除\n- 在共轭梯度法中添加了 Wolfe 搜索\n- 修复了训练算法中的问题\n\n移除：\n- 线性模型\n- Quickprop 训练算法\n- 集成算法","2018-12-02T20:21:02",{"id":191,"version":192,"summary_zh":193,"released_at":194},206255,"v0.6.5","错误修复：\n- 修复了离散霍普菲尔德网络的迭代更新问题：https:\u002F\u002Fgithub.com\u002Fitdxer\u002Fneupy\u002Fissues\u002F218","2018-10-12T13:23:24",{"id":196,"version":197,"summary_zh":198,"released_at":199},206256,"v0.6.4","特性：\n- 新增了生长神经气体（GNG）算法\n- 对基础的GNG算法进行了两项改进\n\n错误修复：\n- 为 `epoch_time` 设置了默认值，以防止在首次迭代时训练被中断而导致程序崩溃。","2018-03-26T06:56:06",{"id":201,"version":202,"summary_zh":203,"released_at":204},206257,"v0.6.3","修复：\n- 使用旧的表格样式，否则在某些字体样式下表格会错位，在 IPython Notebook 中显示效果也不佳。\n- 修复了 GPU 上的 Adam 优化器问题（#200）。","2018-02-03T18:58:47",{"id":206,"version":207,"summary_zh":208,"released_at":209},206258,"v0.6.2","Fixes:\r\n- Fixed neupy installation problem for cases when user don't have pre-installed Theano package (#198)","2017-12-11T13:04:37",{"id":211,"version":212,"summary_zh":213,"released_at":214},206259,"v0.6.1","Enhancement:\r\n- Switched to Theano version 1.0\r\n- Use `tableprint` library instead of neupy table module\r\n- Use `progressbar2` library instead of neupy progressbar module\r\n- Removed logic that controls number of outputs in terminal during the training\r\n\r\nBugs:\r\n- Convert output from the `Step` layer to the integer explicitly in order to avoid boolean outputs\r\n- Fixed issue where message, that tells about training interruption, breaks result table","2017-12-03T19:31:26",{"id":216,"version":217,"summary_zh":218,"released_at":219},206260,"v0.6.0","Features:\r\n* Added module that contains popular DNN architectures, namely Resnet50, VGG19, VGG16, SqueezeNet, AlexNet\r\n* Pre-trained parameters for the new DNN architectures\r\n* Changed format in which neupy stores artitectures.\r\n* To the existed pickle format there were added support for a few new formats, namely hdf5, json and python dictionary\r\n\r\nEnhancement:\r\n* Changed API for the mixture of experts ensemble. Now it works from architectures module.\r\n* Save pickle files using protocol compatible with python 2 and 3 versions\r\n* New error messages that explain failures during parameter loading in storage module\r\n* Use different parameter loading strategies in storage module\r\n\r\nBugs:\r\n* Fixed issue with PNN class mapping (#177)\r\n* Added SOFM weight normalization to the cosine similarity measurement","2017-09-10T15:54:45",{"id":221,"version":222,"summary_zh":223,"released_at":224},206261,"v0.5.2","Features:\r\n* Pickle serializer for networks with fixed architectures\r\n* SOFM weight initialization with PCA\r\n* Added hexagon shaped grid types to SOFM\r\n* Added parameter reduction over time for SOFM\r\n* Possibility to set up different step sizes for different neigbour neurons in SOFM\r\n* Added support for N-dimensional grid shapes for SOFM\r\n* Max-norm regularization algorithm\r\n\r\nEnhancement:\r\n* Made `n_outputs` as an optional parameter for SOFM if `feature_grid` was specified (and vice versa)\r\n\r\nBugs:\r\n* Fixed problem with more complicated cases for inline connections","2017-05-29T20:03:11",{"id":226,"version":227,"summary_zh":228,"released_at":229},206262,"v0.5.1","Features:\r\n* Added LVQ algorithm\r\n* Added LVQ2 algorithm\r\n* Added LVQ2.1 algorithm\r\n* Added LVQ3 algorithm\r\n* Added step reduction algorithm into all LVQ versions\r\n\r\nBugs:\r\n* Changed SciPy version in order to fix problem with golden search algorithm","2017-03-12T22:51:22",{"id":231,"version":232,"summary_zh":233,"released_at":234},206263,"v0.5.0","Features:\n- Added LSTM layer\n- Added GRU layer\n- Added customizable weight and bias initialization for LSTM and GRU layers\n- Created pointer to all layers in the network and connection\n- Ability to extract layer by its name from the connection\n\nEnhancement:\n- Theano 0.9.0 support\n- Fixed issues related to float16 data type\n\nBugs:\n- Modified algorithm for layer name generation\n- Solved problem with Dropout in Wolfe Search algorithm\n- Fixed a few minor bugs\n","2017-02-05T22:27:52",{"id":236,"version":237,"summary_zh":238,"released_at":239},206264,"v0.4.2","Bugs:\n- Fixed input and output layers duplication in the connection\n- Fixed training issues for networks with shared weights\n- Make valid input order for the compile method\n- Fixed progress bar appearance for the training with multiple inputs\n- Fixed shuffle data option for networks with multiple inputs \n","2017-01-10T16:00:54",{"id":241,"version":242,"summary_zh":243,"released_at":244},206265,"v0.4.1","Features:\n- Added Leaky Relu layer\n- Saliency Map plot\n- Ability to specify input and output layers for layer connections (`start` and `end` methods)\n- Added method that compiles network (`compile` method)\n\nEnhancements:\n- Validate that layer name is unique in the network\n- Added ability to train networks with multiple inputs\n","2017-01-04T11:15:25",{"id":246,"version":247,"summary_zh":248,"released_at":249},206266,"v0.4.0","Features:\n- Added Global Pooling layer\n- Added Concatenate layer\n- Added Element-wise layer\n- Added Embedding layer\n- Added Local Response Normalization layer\n- Added Discrete digits dataset\n- Parallel connections\n- Added `layer_structure` plot to visualize relations between layers\n- Added an ability to save and load weights from the pickle file\n\nEnhancement:\n- Improved and modified layer connection API\n- Developed graph structure that stores relations between layers\n- Set up bias as an optional parameter\n- Skip layers for the `layer_structure` function\n- Assign unique identifier for each layer\n\nBugs:\n- Fixed lots of small bug in the layer connection module\n- Fixed bugs with Hessian algorithm\n- Fixed summary table\n","2016-12-11T08:54:16",{"id":251,"version":252,"summary_zh":253,"released_at":254},206267,"v0.3.1","Features:\n- Added Restricted Boltzmann Machine (RBM) (#64)\n\nEnhancement:\n- PNN mini-batch prediction (#72)\n- Check that it's possible to connect two layers during layer connection procedure (#114)\n- Add function for RBM that makes Gibbs sampling from the visible input for multiple iterations (#115)\n- Add more flexible way to initialize network parameters (#110)\n\nRefactoring:\n- Use ParameterProperty class instead of ArrayProperty (#111)\n\nBugs:\n- Fixed Quasi Newton algorithm for the training GPU training (#108)\n- NeuPy shows NaN output values in summary tables as a dash symbol (#109)\n","2016-09-05T17:27:47"]