How iamgopinathbehera/student-performance-indicator works
Project Type and Purpose
• This is a full-stack ML application.
• Offline batch pipeline: ingest raw data, transform, train CatBoost/Scikit-Learn models, store artifacts (*.pkl).
• Online serving: Flask API + simple HTML templates for real-time predictions.
• Deployed via Docker on AWS Elastic Beanstalk (with optional Azure ML integrations).
File Structure
Top-level directories / files:
• .ebextensions + Dockerfile → AWS EB container config
• application.py + templates/ → Flask web service
• src/components/ → data_ingestion.py, data_transformation.py, model_trainer.py
• src/pipeline/ → train_pipeline.py (orchestrates components), predict_pipeline.py (loads artifacts for inference)
• artifacts/ → serialized model.pkl & preprocessor.pkl, CSV datasets
• notebook/ → EDA & training notebooks (for experimentation)
• requirements.txt, setup.py → dependency management
README Insights
• Confirms four major stages: data ingestion, EDA, model training, deployment.
• Tech stack: Python, pandas, scikit-learn, Jupyter, Flask, Docker, AWS, Azure.
Main Components & Their Interactions
a. Offline Pipeline (Batch)
1) Raw Data Source (artifacts/data.csv, notebook/data/stud.csv)
2) DataIngestion (src/components/data_ingestion.py)
3) DataTransformation (src/components/data_transformation.py)
4) ModelTrainer (src/components/model_trainer.py)
5) Artifacts Store (artifacts/model.pkl, preprocessor.pkl, train.csv, test.csv)
b. Online Serving (Real-time)
1) Flask REST API (application.py)
2) PredictPipeline (src/pipeline/predict_pipeline.py) loads model + preprocessor
3) HTML UI (templates/index.html, home.html) calls API
4) Returns prediction to user
c. Deployment Infrastructure
1) Docker Container (Dockerfile)
2) AWS Elastic Beanstalk (.ebextensions/python.config)
3) (Optional) Azure ML or Azure services for retraining or hosting
Architectural Patterns & Design Principles
• Layered / Modular Design: separate components for ingestion, transformation, training, serving.
• Single Responsibility: each Python module handles one step.
• Artifact Management: versioned model + preprocessor in artifacts/
• Reproducibility: notebooks for EDA, pipelines for automation.
• Deployment as immutable container on cloud PaaS.