How azeghaanicet/mlops-project works
Project Type and Purpose
• This is a full-stack MLOps application for real-time Iris flower classification. It combines:
• A modern web frontend (Jinja2 + Tailwind CSS)
• A REST API layer (FastAPI + Pydantic validation)
• An automated ML training pipeline (Scikit-Learn, DVC for data/model versioning)
• Containerized deployment (Docker, docker-compose)
• CI/CD hooks (GitHub Actions) and tests (pytest)
File-Structure Analysis
• Top level:
• app.py, schemas.py → entrypoint for FastAPI + request/response models
• src/train.py → training script
• templates/ → HTML pages (base, index, result, error)
• data/ & models/ → DVC-tracked dataset and serialized model
• metrics.json → stored evaluation metric
• Dockerfile + docker-compose.yml + Makefile + run.ps1 → dev & deployment orchestration
• tests/ → API and model unit tests
• .github/workflows/ci-pipeline.yml → CI pipeline config
• Patterns:
• Layered web app (presentation → API → service/ML → data/versioning)
• MLOps pipeline: data ingestion, training, evaluation, versioning
README Insights
• Lists key endpoints, stack technologies, CI/CD pointers, development vs. production steps
a. Main Components
• User Interface: Jinja2 templates + Tailwind CSS
• API Layer: FastAPI app.py → routes /, /predict, /api/predict, /health, /metrics
• Validation: Pydantic schemas in schemas.py
• ML Model: serialized iris.joblib (LogisticRegression)
• Training Pipeline: src/train.py → data split, model train → metrics.json → DVC
• Data Management: data/iris.csv + DVC for version control of data & model
• Deployment: Docker image & container orchestration via docker-compose
• Support: CI/CD workflow (GitHub Actions), tests (pytest)
b. Relationships & Interactions
• Browser → HTTP GET “/” → UI template rendered
• Browser submits form → HTTP POST “/predict” → FastAPI
• FastAPI → Pydantic → loads model → returns JSON/HTML result
• CLI / CI triggers src/train.py → reads data/iris.csv → trains model → writes metrics.json & iris.joblib → calls DVC to push artifacts
• Dockerfile builds container packaging app, templates, model, dependencies
• docker-compose orchestrates service startup (app, optional sidecars)
c. Architectural Patterns
• Layered architecture: Presentation → Application (API) → Domain (ML service) → Data (DVC storage)
• MLOps pipeline: reproducible training, versioning, model lifecycle
• 12-factor deployment: configuration via environment, containerized, stateless API
d. Key Technologies
• Frontend: Jinja2, TailwindCSS
• API: FastAPI, Uvicorn
• Validation: Pydantic
• ML: Scikit-Learn LogisticRegression
• Data Versioning: DVC
• Containerization: Docker, docker-compose
• Testing: pytest
• CI: GitHub Actions