LOKA x AWS Immersion Day: SageMaker ala Carte

Code used in the AWS Immersion Day: SageMaker ala Carte organized by Loka and celebrated on November 2nd of 2021.

Repo Structure

Repository mainly composed by:

• training
• deployment

1. Training an ML model

In order to show different capabilities of SageMaker when deploying ML models, 2 models were trained prior the event and artifacts were provided to attendants. All code and data related to model training can be found inside training folder.

IMPORTANT NOTE: Before trying the following notebooks, activate the conda environment included in this repo as follows:

conda env create -f environment.yml && conda activate aws-inmmersion

Training a Model with SageMaker's XGBoost

To train a model using xtreme gradient boosting framework that SageMaker uses using SageMaker Python SDK we used the code inside xgboost_customer_churn.ipynb Jupyter notebook. This model is trained using SageMaker instances, therefore, expect a training job to be inititated and an tar.gz artifact to be stored in an S3 bucket in your AWS account.

IMPORTANT NOTE: For executing this one, you should create a proper AWS IAM role with enough permissions to access SageMaker and S3. Read the notebook for further details.

Training a Model with Scikit Learn

To train a model using scikit-learn defacto Python ML library we used the code inside scikitlearn_churn_prediction.py.ipynb Jupyter notebook. This notebook can be executed locally as long as the conda environment in this repo is activated or the Python environment used has scikit-learn==0.24.2.

IMPORTANT NOTE: scikit-learn version should be the same used in [deployment/requirements.txt](requirements.txt) otherwise, deployment will fail.

2. Deploying an ML model

This folder is composed by:

• custom_container: Dockerfile, scripts and configuration files to create a Docker image that ships a scikit-learn model into Elastic Container Registry (ECR) for SageMaker to use when creating an endpoint.
• local_notebook: Notebook to deploy models locally using your AWS CLI credentials/profiles rather than using SageMaker notebooks/SageMaker Studio notebooks.
• model: XGBoost model artifact.
• test_data: Test and validation files to evaluate model performance and send requests to SageMaker endpoint.
• sagemaker-inference-immersion-day-studio-ver-attendee: Version that is meant to be used by the attendees while following instructions given by expositors.
• sagemaker-inference-immersion-day-studio-ver: Version with full code and comments that is meant to be run as it is inside SageMaker Studio.

IMPORTANT NOTE: Workshop notebooks i.e. sagemaker-inference-immersion-day-studio-ver-attendee and sagemaker-inference-immersion-day-studio-ver are meant to be run inside **SageMaker Studio** otherwise, code will fail.