Amazon SageMaker 2026: Benefits, Features & Pricing

I enjoyed the simple machine learning model training using SageMaker. I used it for creating recommender systems for IoT home automation networks. It is very simple to load data sets from with in the Amazon EC2 cloud using the Amazon S3 datastore. The amazing capability in SageMaker is the automatic feature selection and creation for the ML data model using the SageMaker Data Wrangler. This proved to be very accurate and reduced a lot of the noise in my data sets with minimal configuration on my side. Features are even managed in a separate container to be managed later by allowing addition/removal/updating feature entries. Real-time predictions is another great property in Amazon SageMaker in addition to checking the conformity of the data in the ML model across subsets of the data set to ensure balanced data points/features. Sagemaker can even map the role each feature plays in the prediction to give the developer better sense on how the prediction output is biased/affected by each individual feature.

Definitely it would be great to have an offline local reference implementation to test against. However, this is a cloud system after all and accordingly, developers lose the local testing advantages.

J'aime beaucoup passer d'une instance à l'autre, par exemple lorsque je dois faire tourner un modèle de Machine Learning assez lourd avec beaucoup de données, je choisie une instance plus puissante et c'est donc beaucoup plus rapide, cela me fait gagner un temps fou!

Ce que je n'aime pas c'est que les instances s'allument parfois toute seule (et sont parfois payantes), je dois toujours vérifier que seulement celles dont j'ai besoin sont allumées.

I've had fun using Sagemaker to train models, I also enjoyed the functionality that allows us to provision resources to better suit the ML Training needs.

Make it easier to delete domains, since leaving resources on the account incurs charges quickly.