Prediction of Titanic survival: training a classification model locally or on AWS SageMaker, experiment tracking with MLflow, and exposing a FastAPI inference API.
- Load and preprocess Titanic data; train a classifier (e.g. sklearn pipeline).
- Run training locally or as a SageMaker training job (Docker-based).
- Deploy inference as a SageMaker endpoint or run the API locally with uvicorn.
- Track runs and artifacts with MLflow.
src/— data loading, preprocessing, training, evaluation, prediction.api/— FastAPI app for inference.input/— sample data and example JSON input.output/— trained model artifacts and metrics (e.g. after local training).Dockerfile.training,Dockerfile.inference— images for SageMaker.
Copy .env.example (or create .env) at the project root with the following variables:
AWS_REGION=eu-west-1
AWS_ACCOUNT_ID=123456789012
MLFLOW_TRACKING_URI=http://127.0.0.1:5001| Variable | Description |
|---|---|
AWS_REGION |
AWS region used for ECR, S3, and SageMaker |
AWS_ACCOUNT_ID |
AWS account ID (used to build ECR URLs and bucket names) |
MLFLOW_TRACKING_URI |
MLflow tracking server URL — run make run-mlflow-ui to start it locally |
The Makefile derives all other values (AWS_ECR_*_REPOSITORY_URL, AWS_MAIN_BUCKET_NAME, etc.) from these three variables and the project directory name.
Allister K.
MIT License — see LICENSE for details.