Amazon SageMaker is a managed machine learning service. It contains a powerful set of tools that support advanced use cases like training and hyperparameter tuning for custom machine learning models. But SageMaker also provides hosted Jupyter notebook instances, which are a great way to experiment with machine learning in an interactive way. In this tutorial, you will use a notebook to train an RL agent locally and then launch a hyperparameter tuning job.
To launch a SageMaker notebook instance:
Navigate to the Amazon SageMaker console.
Click Create notebook instance.
Give the notebook instance a unique name, like
game-leveling-tutorial. All defaults in Notebook instance settings can be left alone.
In the IAM role drop down menu under Permissions and encryption, choose Create a new role.
Under Specific S3 buckets choose None.
Start the notebook instance by choosing Create notebook instance.
Back in the Notebook instances view of the SageMaker console, the Status will change to ‘InService’ when the notebook instance is ready. Now, choose Open Jupyter.
Start a new Jupyter notebook by clicking on the New dropdown menu and selecting
Give your notebook a name by clicking the title – it should say
Untitled at first. Enter a name that you can remember and choose Rename.
It’s important to open Jupyter and not JupyterLab. This tutorial uses IPython widgets that don’t render correctly in JupyterLab.