内置算法的常见参数 - 亚马逊 SageMaker
AWS 文档中描述的 AWS 服务或功能可能因区域而异。要查看适用于中国区域的差异,请参阅中国的 AWS 服务入门

如果我们为英文版本指南提供翻译,那么如果存在任何冲突,将以英文版本指南为准。在提供翻译时使用机器翻译。

内置算法的常见参数

下表列出了 Amazon SageMaker 提供的每种算法的参数。

算法名称 渠道名称 训练图像和推理图像注册表路径 训练输入模式 文件类型 实例类 可并行化
BlazingText 训练

<ecr_path>/字幕文本:<tag>

文件或管道 文本文件(每行一句,带空格分隔的标记) GPU(仅单个实例) 或 CPU
DeepAR 预测 训练和 (可选) 测试

<ecr_path>/预测-预测:<tag>

File (文件) JSON 行或 Parquet GPU 或 CPU
因子分解机 训练和 (可选) 测试

<ecr_path>/工厂机器:<tag>

文件或管道 recordIO-protobuf CPU(对密集数据使用 GPU)
图像分类 训练和验证,(可选)train_lst、validation_lst 和模型

<ecr_path>/图像分类:<tag>

文件或管道 recordIO 或图像文件 (.jpg 或 .png) GPU
IP 见解 训练和 (可选) 验证

<ecr_path>/透视:<tag>

File (文件) CSV CPU 或 GPU
k-means 训练和 (可选) 测试

<ecr_path>/k平均值:<tag>

文件或管道 recordIO-protobuf 或 CSV CPU或 GPUCommon (一个或多个实例上的单个GPU设备)
k-nearest-neighbor (k-NN) 训练和 (可选) 测试

<ecr_path>/knn(knn):<tag>

文件或管道 recordIO-protobuf 或 CSV CPU 或 GPU(一个或多个实例上的单个 GPU 设备)

LDA

训练和 (可选) 测试

<ecr_path>/lda(lda):<tag>

文件或管道 recordIO-protobuf 或 CSV CPU(仅单个实例)
线性学习器 训练和 (可选) 验证和/或测试 <ecr_path>/线性差:<tag> 文件或管道 recordIO-protobuf 或 CSV CPU 或 GPU
神经主题模型 训练和 (可选) 验证和/或测试

<ecr_path>/元:<tag>

文件或管道 recordIO-protobuf 或 CSV GPU 或 CPU
Object2Vec 训练和 (可选) 验证和/或测试

<ecr_path>/object2vec(对象2vec):<tag>

File (文件) JSON 行 GPU 或 CPU(仅单个实例)
对象检测 训练和验证,(可选)train_annotation、validation_annotation 和模型

<ecr_path>/对象检测:<tag>

文件或管道 recordIO 或图像文件 (.jpg 或 .png) GPU
PCA 训练和 (可选) 测试

<ecr_path>/普卡:<tag>

文件或管道 recordIO-protobuf 或 CSV GPU 或 CPU
随机森林砍伐 训练和 (可选) 测试

<ecr_path>/随机森林:<tag>

文件或管道 recordIO-protobuf 或 CSV CPU
语义分割 训练和验证、train_annotation、validation_annotation 以及(可选)label_map 和模型

<ecr_path>/语义分割:<tag>

文件或管道 图像文件 GPU(仅单个实例)

Seq2Seq 建模

训练、验证和 vocab <ecr_path>/seq2seq(队列2seq):<tag> File (文件) recordIO-protobuf GPU(仅单个实例)
XGBoost 训练和 (可选) 验证

<ecr_path>/xgboost:<tag>

File (文件) CSV或 LibSVM CPU

可并行化 的算法可部署在多个计算实例上以进行分布式训练。对于 Training Image and Inference Image Registry Path (训练镜像和推理镜像仓库路径) 列,可使用 :1 版本标签来确保您使用的是算法的稳定版本。您可以通过在具有 :1 标签的推理镜像上使用具有 :1 标签的镜像来可靠地托管已训练的模型。通过在镜像仓库路径中使用 :latest 标签,可让您获得最新版本的算法,但可能会导致出现与向后兼容性有关的问题。避免将 :latest 标签用于生产用途。

对于 培训图像和推断图像注册路径 列,根据算法和地区,使用以下值之一 <ecr_path>.

算法: BlazingText、图像分类、对象检测、语义分割、 Seq2Seq,和 XGBoost (0.72)(0.72)间

AWS 区域 训练图像和推理图像注册表路径
us-west-1 632365934929.dkr.ecr.us-west-1.amazonaws.com
us-west-2 433757028032.dkr.ecr.us-west-2.amazonaws.com
us-east-1 811284229777.dkr.ecr.us-east-1.amazonaws.com
us-east-2 825641698319.dkr.ecr.us-east-2.amazonaws.com
ap-east-1 286214385809.dkr.ecr.ap-east-1.amazonaws.com
ap-northeast-1 501404015308.dkr.ecr.ap-northeast-1.amazonaws.com
ap-northeast-2 306986355934.dkr.ecr.ap-northeast-2.amazonaws.com
ap-south-1 991648021394.dkr.ecr.ap-south-1.amazonaws.com
ap-southeast-1 475088953585.dkr.ecr.ap-southeast-1.amazonaws.com
ap-southeast-2 544295431143.dkr.ecr.ap-southeast-2.amazonaws.com
ca-central-1 469771592824.dkr.ecr.ca-central-1.amazonaws.com
cn-north-1 390948362332.dkr.ecr.cn-north-1.amazonaws.com.cn
cn-northwest-1 387376663083.dkr.ecr.cn-northwest-1.amazonaws.com.cn
eu-central-1 813361260812.dkr.ecr.eu-central-1.amazonaws.com
eu-north-1 669576153137.dkr.ecr.eu-north-1.amazonaws.com
eu-west-1 685385470294.dkr.ecr.eu-west-1.amazonaws.com
eu-west-2 644912444149.dkr.ecr.eu-west-2.amazonaws.com
eu-west-3 749696950732.dkr.ecr.eu-west-3.amazonaws.com
me-south-1 249704162688.dkr.ecr.me-south-1.amazonaws.com
sa-east-1 855470959533.dkr.ecr.sa-east-1.amazonaws.com
us-gov-west-1 226302683700.dkr.ecr.us-gov-west-1.amazonaws.com

算法: DeepAR 预测

AWS 区域 训练图像和推理图像注册表路径
us-west-1 632365934929.dkr.ecr.us-west-1.amazonaws.com
us-west-2 156387875391.dkr.ecr.us-west-2.amazonaws.com
us-east-1 522234722520.dkr.ecr.us-east-1.amazonaws.com
us-east-2 566113047672.dkr.ecr.us-east-2.amazonaws.com
ap-east-1 286214385809.dkr.ecr.ap-east-1.amazonaws.com
ap-northeast-1 633353088612.dkr.ecr.ap-northeast-1.amazonaws.com
ap-northeast-2 204372634319.dkr.ecr.ap-northeast-2.amazonaws.com
ap-south-1 991648021394.dkr.ecr.ap-south-1.amazonaws.com
ap-southeast-1 475088953585.dkr.ecr.ap-southeast-1.amazonaws.com
ap-southeast-2 514117268639.dkr.ecr.ap-southeast-2.amazonaws.com
ca-central-1 469771592824.dkr.ecr.ca-central-1.amazonaws.com
cn-north-1 390948362332.dkr.ecr.cn-north-1.amazonaws.com.cn
cn-northwest-1 387376663083.dkr.ecr.cn-northwest-1.amazonaws.com.cn
eu-north-1 669576153137.dkr.ecr.eu-north-1.amazonaws.com
eu-central-1 495149712605.dkr.ecr.eu-central-1.amazonaws.com
eu-west-1 224300973850.dkr.ecr.eu-west-1.amazonaws.com
eu-west-2 644912444149.dkr.ecr.eu-west-2.amazonaws.com
eu-west-3 749696950732.dkr.ecr.eu-west-3.amazonaws.com
me-south-1 249704162688.dkr.ecr.me-south-1.amazonaws.com
sa-east-1 855470959533.dkr.ecr.sa-east-1.amazonaws.com
us-gov-west-1 226302683700.dkr.ecr.us-gov-west-1.amazonaws.com

算法: 整流机、IP洞察、k平均值、k最近的相邻、线性学习者、 Object2Vec、神经病学主题模型、PCA和随机森林

AWS 区域 训练图像和推理图像注册表路径
us-west-1 632365934929.dkr.ecr.us-west-1.amazonaws.com
us-west-2 174872318107.dkr.ecr.us-west-2.amazonaws.com
us-east-1 382416733822.dkr.ecr.us-east-1.amazonaws.com
us-east-2 404615174143.dkr.ecr.us-east-2.amazonaws.com
ap-east-1 286214385809.dkr.ecr.ap-east-1.amazonaws.com
ap-northeast-1 351501993468.dkr.ecr.ap-northeast-1.amazonaws.com
ap-northeast-2 835164637446.dkr.ecr.ap-northeast-2.amazonaws.com
ap-south-1 991648021394.dkr.ecr.ap-south-1.amazonaws.com
ap-southeast-1 475088953585.dkr.ecr.ap-southeast-1.amazonaws.com
ap-southeast-2 712309505854.dkr.ecr.ap-southeast-2.amazonaws.com
ca-central-1 469771592824.dkr.ecr.ca-central-1.amazonaws.com
cn-north-1 390948362332.dkr.ecr.cn-north-1.amazonaws.com.cn
cn-northwest-1 387376663083.dkr.ecr.cn-northwest-1.amazonaws.com.cn
eu-central-1 664544806723.dkr.ecr.eu-central-1.amazonaws.com
eu-north-1 669576153137.dkr.ecr.eu-north-1.amazonaws.com
eu-west-1 438346466558.dkr.ecr.eu-west-1.amazonaws.com
eu-west-2 644912444149.dkr.ecr.eu-west-2.amazonaws.com
eu-west-3 749696950732.dkr.ecr.eu-west-3.amazonaws.com
me-south-1 249704162688.dkr.ecr.me-south-1.amazonaws.com
sa-east-1 855470959533.dkr.ecr.sa-east-1.amazonaws.com
us-gov-west-1 226302683700.dkr.ecr.us-gov-west-1.amazonaws.com

算法: 潜在狄利克雷分配 (LDA)

AWS 区域 训练图像和推理图像注册表路径
us-west-1 632365934929.dkr.ecr.us-west-1.amazonaws.com
us-west-2 266724342769.dkr.ecr.us-west-2.amazonaws.com
us-east-1 766337827248.dkr.ecr.us-east-1.amazonaws.com
us-east-2 999911452149.dkr.ecr.us-east-2.amazonaws.com
ap-northeast-1 258307448986.dkr.ecr.ap-northeast-1.amazonaws.com
ap-northeast-2 293181348795.dkr.ecr.ap-northeast-2.amazonaws.com
ap-south-1 991648021394.dkr.ecr.ap-south-1.amazonaws.com
ap-southeast-1 475088953585.dkr.ecr.ap-southeast-1.amazonaws.com
ap-southeast-2 297031611018.dkr.ecr.ap-southeast-2.amazonaws.com
ca-central-1 469771592824.dkr.ecr.ca-central-1.amazonaws.com
eu-central-1 353608530281.dkr.ecr.eu-central-1.amazonaws.com
eu-west-1 999678624901.dkr.ecr.eu-west-1.amazonaws.com
eu-west-2 644912444149.dkr.ecr.eu-west-2.amazonaws.com
us-gov-west-1 226302683700.dkr.ecr.us-gov-west-1.amazonaws.com

算法: XGBoost (0.90年)

AWS 区域 训练图像和推理图像注册表路径
us-west-1 746614075791.dkr.ecr.us-west-1.amazonaws.com
us-west-2 246618743249.dkr.ecr.us-west-2.amazonaws.com
us-east-1 683313688378.dkr.ecr.us-east-1.amazonaws.com
us-east-2 257758044811.dkr.ecr.us-east-2.amazonaws.com
ap-northeast-1 354813040037.dkr.ecr.ap-northeast-1.amazonaws.com
ap-northeast-2 366743142698.dkr.ecr.ap-northeast-2.amazonaws.com
ap-southeast-1 121021644041.dkr.ecr.ap-southeast-1.amazonaws.com
ap-southeast-2 783357654285.dkr.ecr.ap-southeast-2.amazonaws.com
ap-south-1 720646828776.dkr.ecr.ap-south-1.amazonaws.com
ap-east-1 651117190479.dkr.ecr.ap-east-1.amazonaws.com
ca-central-1 341280168497.dkr.ecr.ca-central-1.amazonaws.com
cn-north-1 450853457545.dkr.ecr.cn-north-1.amazonaws.com.cn
cn-northwest-1 451049120500.dkr.ecr.cn-northwest-1.amazonaws.com.cn
eu-central-1 492215442770.dkr.ecr.eu-central-1.amazonaws.com
eu-north-1 662702820516.dkr.ecr.eu-north-1.amazonaws.com
eu-west-1 141502667606.dkr.ecr.eu-west-1.amazonaws.com
eu-west-2 764974769150.dkr.ecr.eu-west-2.amazonaws.com
eu-west-3 659782779980.dkr.ecr.eu-west-3.amazonaws.com
me-south-1 801668240914.dkr.ecr.me-south-1.amazonaws.com
sa-east-1 737474898029.dkr.ecr.sa-east-1.amazonaws.com
us-gov-west-1 414596584902.dkr.ecr.us-gov-west-1.amazonaws.com

使用路径和训练输入模式,如下所示:

  • 要创建训练作业 (使用对 CreateTrainingJob API 的请求),请指定训练镜像的 Docker 镜像仓库路径和训练输入模式。您可以创建训练作业来通过特定数据集训练模型。

     

  • 要创建模型(具有 CreateModel 请求),请指定推理镜像的 Docker 镜像仓库路径。Amazon SageMaker 启动基于终端节点配置的机器学习计算实例并部署模型,其中包括构件(模型训练的结果)。