Deliver your personal ML mannequin into Amazon SageMaker Canvas and generate correct predictions

Machine studying (ML) helps organizations generate income, cut back prices, mitigate danger, drive efficiencies, and enhance high quality by optimizing core enterprise features throughout a number of enterprise items equivalent to advertising and marketing, manufacturing, operations, gross sales, finance, and customer support. With AWS ML, organizations can speed up the worth creation from months to days. Amazon SageMaker Canvas is a visible, point-and-click service that enables enterprise analysts to generate correct ML predictions with out writing a single line of code or requiring ML experience. You should utilize fashions to make predictions interactively and for batch scoring on bulk datasets.

On this submit, we showcase architectural patterns on how enterprise groups can use ML fashions constructed anyplace by producing predictions in Canvas and obtain efficient enterprise outcomes.

This integration of mannequin growth and sharing creates a tighter collaboration between enterprise and knowledge science groups and lowers time to worth. Enterprise groups can use present fashions constructed by their knowledge scientists or different departments to unravel a enterprise drawback as a substitute of rebuilding new fashions in exterior environments.

Lastly, enterprise analysts can import shared fashions into Canvas and generate predictions earlier than deploying to manufacturing with only a few clicks.

Resolution overview

The next determine describes three completely different structure patterns to show how knowledge scientists can share fashions with enterprise analysts, who can then immediately generate predictions from these fashions within the visible interface of Canvas:


To coach and construct your mannequin utilizing SageMaker and produce your mannequin into Canvas, full the next stipulations:

  1. In case you don’t have already got a SageMaker area and Studio person, set up and onboard a Studio user to a SageMaker domain.
  2. Enable and set up Canvas base permissions on your customers and grant users permissions to collaborate with Studio.
  3. It’s essential to have a educated mannequin from Autopilot, JumpStart, or the mannequin registry. For any mannequin that you simply’ve constructed exterior of SageMaker, you should register your mannequin within the mannequin registry earlier than importing it into Canvas.

Now let’s assume the position of an information scientist who’s trying to prepare, construct, deploy, and share ML fashions with a enterprise analyst for every of those three architectural patterns.

Use Autopilot and Canvas

Autopilot automates key duties of an automated ML (AutoML) course of like exploring knowledge, deciding on the related algorithm for the issue sort, after which coaching and tuning it. All of this may be achieved whereas permitting you to keep up full management and visibility on the dataset. Autopilot robotically explores completely different options to seek out the most effective mannequin, and customers can both iterate on the ML mannequin or immediately deploy the mannequin to manufacturing with one click on.

On this instance, we use a buyer churn artificial dataset from the telecom area and are tasked with figuring out prospects which can be doubtlessly prone to churning. Full the next steps to make use of Autopilot AutoML to construct, prepare, deploy, and share an ML mannequin with a enterprise analyst:

  1. Obtain the dataset, add it to an Amazon S3 (Amazon Simple Storage Service) bucket, and make an observation of the S3 URI.
  2. On the Studio console, select AutoML within the navigation pane.
  3. Select Create AutoML experiment.
  4. Specify the experiment title (for this submit, Telecom-Buyer-Churn-AutoPilot), S3 knowledge enter, and output location.
  5. Set the goal column as churn.
  6. Within the deployment settings, you may allow the auto deploy choice to create an endpoint that deploys your finest mannequin and runs inference on the endpoint.

For extra data, discuss with Create an Amazon SageMaker Autopilot experiment.

  1. Select your experiment, then choose your finest mannequin and select Share mannequin.
  2. Add a Canvas person and select Share to share the mannequin.

(Notice: You may’t share mannequin with the identical Canvas person as used for Studio login. For instance, Studio user-A can’t share mannequin with Canvas Consumer-A. However user-A can share mannequin with user-B, therefore select completely different makes use of for model-sharing)

For extra data, discuss with Studio users: Share a model to SageMaker Canvas.

Use JumpStart and Canvas

JumpStart is an ML hub that gives pre-trained, open-source fashions for a variety of ML use circumstances like fraud detection, credit score danger prediction, and product defect detection. You may deploy greater than 300 pre-trained fashions for tabular, imaginative and prescient, textual content, and audio knowledge.

For this submit, we use a LightGBM regression pre-trained mannequin from JumpStart. We prepare the mannequin on a customized dataset and share the mannequin with a Canvas person (enterprise analyst). The pre-trained mannequin may be deployed to an endpoint for inference. JumpStart gives an instance pocket book to entry the mannequin after it’s deployed.

On this instance, we use the abalone dataset. The dataset accommodates examples of eight bodily measurements equivalent to size, diameter, and peak to foretell the age of abalone (a regression drawback).

  1. Obtain the abalone dataset from Kaggle.
  2. Create an S3 bucket and add the prepare, validation, and customized header datasets.
  3. On the Studio console, below SageMaker JumpStart within the navigation pane, select Fashions, notebooks, options.
  4. Beneath Tabular Fashions, select LightGBM Regression.
  5. Beneath Practice Mannequin, specify the S3 URIs for the coaching, validation, and column header datasets.
  6. Select Practice.
  7. Within the navigation pane, select Launched JumpStart property.
  8. On the Coaching jobs tab, select your coaching job.
  9. On the Share menu, select Share to Canvas.
  10. Select the Canvas customers to share with, specify the mannequin particulars, and select Share.

For extra data, discuss with Studio users: Share a model to SageMaker Canvas.

Use SageMaker mannequin registry and Canvas

With SageMaker mannequin registry, you may catalog fashions for manufacturing, handle mannequin variations, affiliate metadata, handle the approval standing of a mannequin, deploy fashions to manufacturing, and automate mannequin deployment with CI/CD.

Let’s assume the position of an information scientist. For this instance, you’re constructing an end-to-end ML mission that features knowledge preparation, mannequin coaching, mannequin internet hosting, mannequin registry, and mannequin sharing with a enterprise analyst. Optionally, for knowledge preparation and preprocessing or postprocessing steps, you should use Amazon SageMaker Data Wrangler and an Amazon SageMaker Processing job. On this instance, we use the abalone dataset downloaded from LIBSVM. The goal variable is the age of abalone.

  1. In Studio, clone the GitHub repo.
  2. Full the steps listed within the README file.
  3. On the Studio console, below Fashions within the navigation pane, select Mannequin registry.
  4. Select the mannequin sklearn-reg-ablone.
  5. Share mannequin model 1 from the mannequin registry to Canvas.
  6. Select the Canvas customers to share with, specify the mannequin particulars, and select Share.

For directions, discuss with the Mannequin Registry part in Studio users: Share a model to SageMaker Canvas.

Handle shared fashions

After you share the mannequin utilizing any of the previous strategies, you may go to the Fashions part in Studio and evaluation all shared fashions. Within the following screenshot, we see 3 completely different fashions shared by a Studio person (knowledge scientist) with completely different Canvas customers (enterprise groups).

Import a shared mannequin and make predictions with Canvas

Let’s assume the position of enterprise analyst and log in to Canvas along with your Canvas person.

When an information scientist or Studio person shares a mannequin with a Canvas person, you obtain a notification inside the Canvas utility {that a} Studio person has shared a mannequin with you. Within the Canvas utility, the notification is much like the next screenshot.

You may select View replace to see the shared mannequin, or you may go to the Fashions web page within the Canvas utility to find all of the fashions which were shared with you. The mannequin import from Studio can take as much as 20 minutes.

After importing the mannequin, you may view its metrics and generate real-time predictions with what-if analysis or batch predictions.


Take note the next when sharing fashions with Canvas:

  • You retailer coaching and validation datasets in Amazon S3, and the S3 URIs are handed to Canvas with AWS Identity and Access Management (IAM) permissions.
  • Present the goal column to Canvas or use the primary column as default.
  • For a Canvas container to parse inference knowledge, the Canvas endpoint accepts both textual content (CSV) or utility (JSON).
  • Canvas doesn’t help a number of container or inference pipelines.
  • An information schema is supplied to Canvas if no headers are supplied within the coaching and validation datasets. By default, the JumpStart platform doesn’t present headers within the coaching and validation datasets.
  • With Jumpstart, the coaching job must be full earlier than you may share it with Canvas.

Discuss with Limitations and troubleshooting that will help you troubleshoot any points you encounter when sharing fashions.

Clear up

To keep away from incurring future fees, delete or shut down the sources you created whereas following this submit. Discuss with Logging out of Amazon SageMaker Canvas for extra particulars. Shut down the person sources, together with notebooks, terminal, kernels, apps and situations. For extra data, discuss with Shut Down Resources. Delete the model version, SageMaker endpoint and resources, Autopilot experiment resources, and S3 bucket.


Studio permits knowledge scientists to share ML fashions with enterprise analysts in a couple of easy steps. Enterprise analysts can profit from ML fashions already constructed by knowledge scientists to unravel enterprise issues as a substitute of making a brand new mannequin in Canvas. Nevertheless, it could be troublesome to make use of these fashions exterior the environments through which they’re constructed attributable to technical necessities and handbook processes to import fashions. This typically forces customers to rebuild ML fashions, ensuing within the duplication of effort and extra time and sources. Canvas removes these limitations so you may generate predictions in Canvas with fashions that you’ve got educated anyplace. Through the use of the three patterns illustrated on this submit, you may register ML fashions within the SageMaker mannequin registry, which is a metadata retailer for ML fashions, and import them into Canvas. Enterprise analysts can then analyze and generate predictions from any mannequin in Canvas.

To be taught extra about utilizing SageMaker companies, take a look at the next sources:

In case you have questions or ideas, depart a remark.

Concerning the authors

Aman Sharma is a Senior Options Architect With AWS. He works with start-ups, small and medium companies, and enterprise prospects throughout the APJ area, greater than 19 years of expertise in consulting, architecting, and solutioning. He’s obsessed with democratizing AI and ML and serving to prospects in designing their knowledge and ML methods. Exterior work, he likes to discover nature and wildlife.

Zichen Nie is the Senior Software program Engineer at AWS SageMaker main the mission Deliver Your Personal Mannequin to SageMaker Canvas final 12 months. She has been working in Amazon for greater than 7 years and has expertise in each Amazon Provide Chain Optimization and AWS AI companies. She enjoys Barre exercises and music after work.

Leave a Reply

Your email address will not be published. Required fields are marked *