Training Jobs

Once data has been uploaded, users can review navigate to the training jobs page within a project. The training jobs page will list completed and in-progress jobs as well as allow users to take various actions against those jobs including creating new jobs, viewing details for existing jobs and stopping running jobs.

Create a Training Job

From the training jobs dashboard users can create new training jobs by selecting the resources to allocate, choosing an algorithm, providing hyperparameters, identify the data channels, and submitting a training job. Amazon SageMaker, on the backend, provisions the resources required for training, and simply lets the user know when it’s done!

Custom Algorithm

When creating training jobs you can choose from a built-in alogorithm or to bring your own algorithm from ECR. When using a custom algorithm you will need to define metric definitions as well as hyper parameters when creating the training job. Using a custom algorithm will require that MLSpace has been configured to allow access to ECR.

Training Job details

The details view for training jobs allows you to review the settings used to create the job including, the algorithm, hyperparameters, input data, and the dataset/path where the training job results will be stored.

The "Create model" button will lead to the create model form and pre-populate it with the algorithm and training data output folder from the training job.

The details view also allows you to view the logs associated with the training job run.