AWS Marketplace
User Guide for AWS Marketplace Subscribers

Machine Learning Products

AWS Marketplace has a category for machine learning products you can subscribe to through AWS Marketplace. The product category is Machine Learning. The products in this category include machine learning (ML) algorithms and model packages.

An Amazon SageMaker algorithm is a unique Amazon SageMaker entity that is identified by an Amazon Resource Name (ARN). An algorithm has two logical components: training and inference. Customers use the training component to create a training job or tuning job using your input dataset in Amazon SageMaker to build machine learning models. Amazon SageMaker saves the model artifacts generated by the algorithm during training to an S3 bucket. Customers can build a model package using the algorithm’s inference component and the model artifacts that are stored in the S3 bucket. Customers can use this model package to build a model, which can then be used for running on hosting services or running batch transforms in Amazon SageMaker.

An Amazon SageMaker model package is a unique pretrained ML model that is identified by an ARN on Amazon SageMaker. Customers use a model package to create a model in Amazon SageMaker, which can then be used with hosting services to run real-time inference or with batch transform to run batch inference in Amazon SageMaker.

You can browse and search for hundreds of machine learning algorithms and model packages from a broad range of subcategories, such as computer vision, natural language processing, speech recognition, text, data, voice, image, video analysis, fraud detection, and predictive analysis.

To assess the quality and suitability of a model, you can review product descriptions, usage instructions, customer reviews, sample Jupyter notebooks, pricing, and support information. You deploy models directly from the Amazon SageMaker console, through a Jupyter notebook, with the Amazon SageMaker SDK, or using the AWS Command Line Interface (AWS CLI). Amazon SageMaker provides a secure environment to run your training and inference jobs by running a static scan on all marketplace products.

Find, Subscribe, and Deploy

To find, buy and deploy machine learning products, you find and subscribe to products on AWS Marketplace and then deploy the product on Amazon SageMaker.

You pay only for your usage, with no minimum fees or upfront commitments. AWS Marketplace provides a consolidated bill for algorithm and model package, and AWS infrastructure usage charges. To find, subscribe, and deploy Amazon SageMaker algorithms and model packages:

  1. From the AWS Marketplace website, under Find AWS Marketplace products that meet your needs, use the Categories pull-down menu to find the subcategory under Machine Learning that you are interested in. A full list of categories can be found in the Product Categories section of this user guide. You can refine your search results by applying resource type, category, and pricing filters. From search results, you can access the product detail page, which allows you to review the product description, usage instructions, customer reviews, data requirements, sample Jupyter notebooks, and pricing and support information.

  2. To view the procurement page, from the product detail page, choose Continue to subscribe. After reviewing the product pricing information and the end user license agreement (EULA), you can subscribe. After subscribing, you can configure the product (for example, by selecting a specific version or deployment region) on the AWS Marketplace website.

  3. After configuring the product, you can view the Amazon SageMaker product detail page by choosing View in Amazon SageMaker. From the Amazon SageMaker console, you can deploy the algorithms and model packages using the Amazon SageMaker console, Jupyter notebook, Amazon SageMaker CLI commands, or API operations.

To deploy a third-party algorithm/model package on Amazon SageMaker, you need a valid subscription. Find the suitable algorithm/model package from AWS Marketplace and subscribe to the products. Navigate to Your Marketplace Software and make sure that you have a valid subscription to the algorithm you want to deploy.

For more information on deploying on Amazon SageMaker, see Getting Started.

Frequently Asked Questions

I am new to Amazon SageMaker. How can I learn how to use it?

We have resources to help you to get started with Amazon SageMaker. You can read the Amazon SageMaker documentation, the API reference, the build-train-deploy tutorial, and pricing information. In addition, you can watch the following webinars:

Where can I find the third-party algorithms and models?

After signing in as a whitelisted account user, you can access the preconfigured URL to find algorithms and models that can be deployed on Amazon SageMaker. You can also find the algorithms and models by navigating to AWS Marketplace, searching for the keyword “algorithm” or “model,” and filtering by the “Amazon SageMaker” delivery method from the search filters in the left navigation.

What kind of products are available on AWS Machine Learning Marketplace?

AWS Machine Learning Marketplace offers a wide range of ML algorithms and model packages from a broad range of categories, such as text, data, voice, image, video analysis, fraud detection, predictive analysis, and more. It also includes industry-specific ML products (demand forecasting, predicting engagement, etc.) for the financial services, healthcare, media and entertainment, oil and gas, information technology, manufacturing, and telecom industries.

How much does it cost to use algorithms or model packages?

Algorithm and model package price consists of an hourly usage fee set by the seller and an infrastructure fee billed per hour based on the Amazon SageMaker resource usage. For the algorithm or model package fee, sellers define the hourly fee for each supported instance type. When buying algorithms, you will see the training price for the algorithm and the inference price for the model derived from the algorithm. When buying model packages, you will see only the inference price. For free products or products in free-trial offers, you will be charged only for Amazon SageMaker resource usage (there are no seller charges).

Why are there two different prices (training and inference) for an algorithm?

You pay the training price when you train a model using the seller's algorithm. To run inference with the model, you use the seller’s inference image and pay the inference price for that.

Can I use my own inference logic for inference?

If the seller has not encrypted the model artifacts, you can use your own inference logic. Amazon does not require sellers to encrypt model artifacts. However, sellers can choose to encrypt model artifacts to protect their IP. If the seller has encrypted the model artifacts, you can request the encryption keys from the seller or use the seller’s inference image for inference.

How safe is the data that I use to train or run inference with an algorithm or model package?

Amazon SageMaker provides a secure environment for training your data or running inference with third-party algorithms or models. To ensure data security, it takes the following measures:

  • Amazon SageMaker performs static scans for security vulnerabilities of all third-party algorithms and model packages.

  • Amazon SageMaker encrypts algorithms and model artifacts and other system artifacts in transit and at rest.

  • Requests to Amazon SageMaker APIs and the console are made over a secure (SSL) connection.

  • Amazon SageMaker requires AWS Identity and Access Management (AWS IAM) credentials to access your deployment’s resources and data. This prevents sellers from accessing your data.

How can I get support for my purchase from the seller?

The level of product support is determined by the seller. It can range from direct support by phone or email to information provided on the seller website. To contact the seller, use the contact information listed on the product detail page.

What reports do I get about my service usage?

You can get detailed billing reports for the products that you used in the AWS Billing and Cost Management console. Reports provide information about the products in use, number of hours that you have used each, and the charges for each algorithm or model that you have used. Usage reports are provided at daily, weekly, and monthly intervals. You can also get Amazon CloudWatch metrics, AWS CloudTrail logs, and Amazon CloudWatch Logs log files for your deployments from the Amazon SageMaker console.