AWS Machine Learning Blog

Amazon SageMaker console now supports training job cloning

Today we are launching the training job cloning feature on the Amazon SageMaker console, which makes it much easier for you to create training jobs based on existing ones.

When you use Amazon SageMaker, it’s common to run multiple training jobs using different training sets and identical configuration. It’s also common to adjust a specific hyperparameter, or the underlying compute resources, and train a model again. In such scenarios, creating new training jobs based on existing ones would be preferred. To address this requirement, Amazon SageMaker now supports the cloning of existing training jobs, using the AWS Management Console.

Cloning an existing training job using the console is easy. On the Amazon SageMaker console, select Training, then Training jobs on the left navigation pane. Then you will see a list of all the training jobs that you have created so far.

Let’s say you want to create a new training job. We’ll use the job that is currently in the Training jobs list (xgboost-2018-06-05-17-19-32-703) as an example. You can use the same settings to train xgboost-2018-06-05-17-19-32-703 against a different dataset. Instead of choosing Create training job to create a training job from scratch, you simply check xgboost-2018-06-05-17-19-32-703 and choose Clone to open the Clone training job page. 

Notice that the job settings look almost the same as the original xgboost-2018-06-05-17-19-32-703, except that a suffix was automatically added to the original job name. If you want to rerun the same training process you can just choose Create training job at the bottom of the page without changing anything. Since we are going to use a different dataset, you can rename the training job name as you like. In this example we use xgboost-mydataset. Scroll down to Input data configuration. Here you can change the Amazon S3 location to where your new dataset is, then choose Done. 

Now you are ready to go! Just scroll down to the bottom of the page, choose Create training job, and the next thing you know is that the new training job is up and running!

You can also clone a training job on the training job detail page, since in some cases you might want to check the detail settings of a training job before cloning it. The workflow is the same after you choose Clone on the job detail page.

Training job cloning through the Amazon SageMaker console is available today in the US East (N. Virginia), US East (Ohio), EU (Ireland), U.S. West (Oregon), and Asia Pacific (Tokyo) AWS Regions. To learn more, visit Amazon SageMaker console.

About the Author

Fan Li is a Product Manager in the AWS ML Platforms team, which includes Amazon SageMaker, Amazon Machine Learning, and the AWS Deep Learning AMIs. He used to be a big fan of ballroom dance but now loves whatever his 7-year-old son likes.