{"id":19196455,"url":"https://github.com/lingdu2012/paddlenote","last_synced_at":"2026-02-03T00:02:14.420Z","repository":{"id":124835279,"uuid":"359478021","full_name":"lingdu2012/paddleNote","owner":"lingdu2012","description":"百度飞桨学习笔记","archived":false,"fork":false,"pushed_at":"2021-04-19T14:12:00.000Z","size":2,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-02-23T04:44:01.729Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/lingdu2012.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2021-04-19T13:55:05.000Z","updated_at":"2021-04-19T14:12:02.000Z","dependencies_parsed_at":null,"dependency_job_id":"e6a57a74-1ddc-4f6c-8407-00688b1c3751","html_url":"https://github.com/lingdu2012/paddleNote","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/lingdu2012/paddleNote","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lingdu2012%2FpaddleNote","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lingdu2012%2FpaddleNote/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lingdu2012%2FpaddleNote/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lingdu2012%2FpaddleNote/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/lingdu2012","download_url":"https://codeload.github.com/lingdu2012/paddleNote/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lingdu2012%2FpaddleNote/sbom","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":261462685,"owners_count":23162012,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-11-09T12:13:44.602Z","updated_at":"2026-02-03T00:02:09.400Z","avatar_url":"https://github.com/lingdu2012.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"# paddleNote\n百度飞桨学习笔记\n\n# 基础知识（一）\n\n* 人工智能 \u003e 机器学习 \u003e 深度学习\n\n* 机器学习的实现可以分成两步：训练（归纳：从具体案例中抽象一般规律）和预测（演绎：从一般规律推导出具体案例的结果）\n\n* 模型有效的基本条件是能够拟合已知的样本\n\n* 衡量模型预测值和真实值差距的评价函数也被称为损失函数（损失Loss）\n\n* 最小化损失是模型的优化目标，实现损失最小化的方法称为优化算法，也称为寻解算法（找到使得损失函数最小的参数解）。\n\n* 模型假设、评价函数（损失/优化目标）和优化算法是构成模型的三个部分。\n\n* 机器执行学习的框架体现了其学习的本质是“参数估计”（Learning is parameter estimation）\n\n* 机器学习和深度学习在理论结构上是一致的，即：模型假设、评价函数和优化算法，其根本差别在于假设的复杂度\n\n* 神经元： 神经网络中每个节点称为神经元，由两部分组成：\n  加权和：将所有输入加权求和。\n  非线性变换（激活函数）：加权和的结果经过一个非线性函数变换，让神经元计算具备非线性的能力。\n  多层连接： 大量这样的节点按照不同的层次排布，形成多层的结构连接起来，即称为神经网络。\n  前向计算： 从输入计算输出的过程，顺序从网络前至后。\n  计算图： 以图形化的方式展现神经网络的计算逻辑又称为计算图。\n\n* 神经网络并没有那么神秘，它的本质是一个含有很多参数的“大公式”。\n\n* SIFT，即尺度不变特征变换（Scale-invariant feature transform，SIFT），是用于图像处理领域的一种描述。这种描述具有尺度不变性，可在图像中检测出关键点，是一种局部特征描述子。\n\n* 神经网络的标准结构中每个神经元由加权和与非线性变换构成，然后将多个神经元分层的摆放并连接形成神经网络。\n\n  \n\n# 基础知识（二）\n\n* 构建神经网络/深度学习模型的基本步骤：数据处理、模型设计、训练配置、训练过程、模型保存\n* 数据处理：数据导入、数据形状变换、数据集划分、数据归一化处理、封装load data 函数\n* 数据集：训练集（确定模型的参数）和测试集（评判模型的效果）\n* 数据归一化处理：使每个特征的取值缩放到0-1之间，可以使模型训练更高效，特征前的权重大小可以代表该变量对预测结果的贡献度\n* 将数据处理的操作封装成 load data 函数有助于模型的调用\n* 对于回归问题，最常采用的衡量方法是使用均方误差作为评价模型好坏的指标；对于分类问题，通常采用交叉熵作为损失函数\n* Loss=(y−z)2\n* 模型训练的目标是让定义的损失函数尽可能的小，即取得极小值\n* 基于Numpy广播机制，可以更快速地实现梯度计算\n* 学习率：控制每次参数值沿着梯度反方向变动的大小，即每次移动的步长。\n* 随机梯度下降法（Stochastic Gradient Descent，SGD）：每次从总的数据集中随机抽取出小部分数据来代表整体，基于这部分数据计算梯度和损失来更新参数。\n* 数据处理需要实现拆分数据批次和样本乱序（为了实现随机抽样的效果）两个功能。\n* 当前大部分深度学习框架在建模过程中，每一步都可以分为两个部分：个性化（指定模型由哪些逻辑元素组合，由建模者完成）和通用化（聚焦这些元素的算法实现，由框架完成）","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Flingdu2012%2Fpaddlenote","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Flingdu2012%2Fpaddlenote","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Flingdu2012%2Fpaddlenote/lists"}