AWS launches ML dev service in Sydney region

Amazon Web Services (AWS) has launched Amazon SageMaker Studio, an integrated development environment for machine learning (ML), in its Sydney region. 

The service allows users to build, train and deploy ML models through a web-based visual interface for AWS' SageMaker solution. 

Broadly, the service allows for the upload of data, new notebook creation, model training and tuning, experiment adjustment, result comparisons, and model deployment.

AWS claims ML development is covered by SageMaker throughout the whole process, with notebook creation covered by SageMaker Studio Notebooks, Autopilot for automatic model building, training and tuning and Ground Truth for training data labelling.  

It also includes Experiments for ML model iteration organisation and tracking, Debugger for analysing and debugging anomalies, Managed Spot Training for cost reduction and Model Monitor for the detection and remediation of concept drift. 

The service also supports deep learning frameworks from TensorFlow, PyTorch, Apache MXNet, Chainer, Keras, Gluon, Horovod, Scikit-learn, and Deep Graph Library. 

Amazon SageMaker Studio itself is free, but users have to pay for the AWS services used within Studio. 

However, Some SageMaker services are available for free under AWS’ free tier. For those that have not used SageMaker before, users will gain limited access to two months’ worth of free notebook usage, model training and deployment. 

This includes 250 hours per month of t2.medium or t3.medium notebook usage with on-demand notebook instances or t3.medium instances with SageMaker Studio notebooks for model building.  

For model training, the free tier has 50 hours of m4.xlarge or m5.xlarge. 

Meanwhile, the free tier has 125 hours of m4.xlarge or m5.xlarge for ML models for real-time inference and batch transform with Amazon SageMaker.  

However, the free tier does not cover storage volume usage. As for the paid tier, the cost of Studio Notebooks, on-demand notebooks, processing, training, real-time inference and batch transform varies depending on usage, but is charged by the hour.