- Business Needs
- EDA
- Model Building in Google Colab using PyCaret
- Experiment Tracking using MLflow
- Configuration Tracking
- Building a web front-end to serve the Model using Streamlit
- Deploying the Streamlit app to Heroku, Streamlit Share and Docker
This was assignment #3 in DPhi Tech's Machine Learning Bootcamp Advanced track. I enrolled in the bootcamp to sharpen the saw for a variety of skills. Initially I built out a simple ML pipeline using pandas and sklearn and restarted the assignment using PyCaret, informed by the EDA and restricted evaluation from my simple pipeline. This gave me a greater appreciation for what PyCaret brings to ML and MLOps workflows that can be used by Citizen Data Scientists, Machine Learning Engineers and Data Scientists. From my prior experience working with Azure Machine Learning, I have started the initial iteration to build capabilities that a data scientist would need when working alongside DBA, MLOps and Full-Stack Web Developers in a SCRUM/Product-Centric team in order to capture salient information through the DTAP lifecycle.
Heroku Note: The current implementation and dependencies results in a slug size of 574MB which exceeds the max slug size (500MB) of the free Heroku tier dyno.
Streamlit Share https://share.streamlit.io/salilathalye/cwa-dphitech-challenge-54/main/src/app.py
DPhi Tech Challenge #54 https://dphi.tech/practice/challenge/54#data