SageMaker Fridays Season 3, Episode 3 - Managing engineered features with SageMaker Feature Store

⭐️⭐️⭐️ Don't forget to subscribe to be notified of future videos ⭐️⭐️⭐️

In this episode, we build a sentiment analysis model starting from the Amazon Customer Reviews dataset. First, we import the dataset in Parquet format in Amazon Athena. Then, we import it from Athena to SageMaker Data Wrangler for a quick look. Then, we move to a Jupyter notebook and we start engineering features using popular open source libraries (nltk and spaCy), and we automate them with SageMaker Processing. Next, we load the processed dataset in SageMaker Feature Store, both offline and online. Next, we run Athena queries on the offline store in order to build a training set, which we use to train and deploy a sentiment analysis model with the built-in BlazingText algorithm. Finally, we see how to update and delete individual features in the online store, and how to use timestamps for feature versioning.

100% live, no slides :)

[juliensimonfr]

00:00 Welcome and introduction
08:20 The Amazon Customer Reviews dataset
09:45 Importing the dataset into Amazon Athena
12:30 Importing the dataset into SageMaker Data Wrangler
15:05 Engineering features in SageMaker Studio
23:30 Automating feature engineering with SageMaker Processing
29:00 Importing features into SageMaker Feature Store
37:55 Building a dataset with Amazon Athena
44:10 SageMaker business as usual: training and deploying our model
49:10 Working with the online feature store, and traveling time
57:50 Wrap up and useful resources

[jaysoni]

Hey thanks for detail video,
I understood offline feature store data stored in s3 but how about online feature store, where data going to store?

[MuhammadAli-migg]

HI Julien, thanks a lot for your book. I am learning from it on Packt.

But in chap 14, when I am trying to save a tensorflow model, to use it for deploying it.
I can't figure out, why model is not being saved in the 'model_dir'. I have seen it is written that we have to save the model in 'opt/ml/model'..

Please, hep me out with