Sentiment Analysis - MLOps Capstone Project

This repository demonstrates a full-scale MLOps pipeline for a Sentiment Analysis model trained on the IMDB movie review dataset. It combines best practices from software engineering and machine learning to create a robust, reproducible, and deployable ML system.

---

Tools & Technologies Used

Category	Tools/Services Used
Version Control	Git, GitHub
Virtual Env	Conda
Experiment Tracking	MLFlow (via Dagshub)
Data Versioning	DVC
Model Serving	Flask API
CI/CD	GitHub Actions
Containerization	Docker, DockerHub
Cloud Storage	AWS S3 (for DVC remote), AWS ECR (for container image storage)

---

Setup Instructions

1. Initial Setup

git clone https://github.com/AdArya125/MLOPS-Capstone-Project.git && cd MLOPS-Capstone-Project
conda create -n atlas python=3.10 -y
conda activate atlas
pip install cookiecutter dagshub mlflow dvc dvc[s3] awscli flask pipreqs

2. Project Bootstrapping

cookiecutter -c v1 https://github.com/drivendata/cookiecutter-data-science
mv src/models src/model

---

Experiment Tracking with Dagshub & MLFlow

1. Create a Dagshub repo and connect your GitHub repo.
2. Copy the MLFlow tracking URI and auth token.
3. Run experiment notebooks and track metrics via MLFlow.

from dagshub import dagshub_logger
import mlflow

mlflow.set_tracking_uri("https://dagshub.com/<user>/<repo>.mlflow")
mlflow.set_experiment("sentiment-analysis")

with mlflow.start_run():
    ...

---

DVC Pipeline Setup

dvc init
mkdir local_s3
dvc remote add -d mylocal local_s3

• Implement components inside src/:

  • logger.py
  • data_ingestion.py
  • data_preprocessing.py
  • feature_engineering.py
  • model_building.py
  • model_evaluation.py
  • register_model.py

• Add dvc.yaml and params.yaml.

dvc repro
dvc status
git add .
git commit -m "DVC and MLFlow tracking setup"
git push

---

Cloud Storage with AWS S3

aws configure
dvc remote add -d myremote s3://<your-bucket-name>

Ensure your IAM user has AmazonS3FullAccess.

---

Flask API for Inference

cd flask_app
pip install flask
python app.py

---

CI/CD Pipeline (GitHub Actions)

• Add .github/workflows/ci.yaml
• Generate Dagshub token and save it in GitHub secrets as CAPSTONE_TEST
• Add AWS-related secrets:

AWS_ACCESS_KEY_ID
AWS_SECRET_ACCESS_KEY
AWS_REGION
AWS_ACCOUNT_ID
ECR_REPOSITORY

---

Docker Setup

pipreqs flask_app --force
docker build -t capstone-app:latest .
docker run -p 8888:5000 -e CAPSTONE_TEST=your_token capstone-app:latest

(Optional):

docker push youruser/capstone-app:latest

---

AWS ECR Deployment (Optional)

1. Create ECR repo in AWS.
2. Ensure IAM user has AmazonEC2ContainerRegistryFullAccess.
3. Add ECR build/push steps to CI pipeline.

---

Testing

Use the tests/ and scripts/ directories for unit tests.

pytest tests/

Acknowledgements

• DrivenData's Cookiecutter Template
• Dagshub
• AWS S3 & ECR
• DVC
• MLFlow

---

Contact

Aditya Arya 📧 [email protected] 🌐 LinkedIn 🔗 GitHub

---

License

This project is licensed under the MIT License. See the LICENSE file for details.