Today, at AWS re: Invent in Las Vegas, the company announced AutoPilot, a new tool that gives you more insight into the automatic creation of machine learning models called AutoML. This new tool is part of today also announced new SageMaker Studio.
As Andy Jassy, CEO of AWS, said today on stage, one of the problems with AutoML is that it's basically a black box. If you want to improve a mediocre model or just develop it for your business, you do not know how it was created.
The idea behind AutoPilot is to give you the ease of modeling that you get with an AutoML-generated model, but also get a deeper understanding of how the system created the model. "AutoPilot is a way to automatically create a model, but gives you complete visibility and control," said Jassy.
Dataset and performs a series of candidates to determine the optimal combination of data preprocessing steps, machine learning algorithms, and hyperparameters. Then, this combination is used to train an inference pipeline, which you can easily deploy either on a real-time endpoint or for batch processing. As with Amazon SageMaker, it's all done on a fully managed infrastructure, "the company said in a blog post announcing the new feature.
You can view the parameters of the model, and 50 automated models show you a ranking of which models performed best. You can also look at the model's underlying notebook and see what compromises have been made to generate the best model. For example, it is the most accurate, but it brings down speed.
Your organization may have its own requirements, and you can choose the best model based on the parameters that you consider most important. It was generated in an automated way.
Once you have the model you like best, you can select it in SageMaker Studio and launch it with a single click. The tool is available now.