MLflow Model Registry Application

A comprehensive full-stack machine learning application featuring pose detection, exercise analysis, and performance scoring. The system uses models registered in MLflow for expert score prediction, weakest link classification, and real-time squat form analysis through pose estimation.

Note: For an in-depth explanation of our system architecture, coding conventions, testing strategies, API design, security practices, and development workflow, please refer to our GitLab Wiki (out dated):
Design-and-Coding-Rules-Summary

---

🚀 Quick Start

# 1. Start MLflow Server
mlflow server --host 127.0.0.1 --port 4800

# 2. Set up Backend
cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt
python src/app.py

# 3. Set up Frontend
cd frontend
npm install
npm run build
npm run preview

Visit:

• Frontend: http://localhost:4173
• Backend: http://localhost:5000
• MLflow UI: http://127.0.0.1:4800

---

📋 Requirements

Backend

• Python 3.11
• Key dependencies (all included in requirements.txt):
  • FastAPI 0.115.8
  • MLflow 2.20.3
  • uvicorn 0.34.0
  • python-multipart 0.0.20
  • TensorFlow 2.19.0
  • Keras 3.9.2
  • pandas 2.2.3
  • joblib 1.4.2
  • aiofiles 23.2.1

Frontend

• Node.js 22.13.1
• npm 10.9.2
• Vue.js 3.5.13
• Key dependencies:
  • @tensorflow/tfjs 4.13.0
  • @tensorflow-models/pose-detection 1.0.0
  • three.js 0.176.0
  • axios 1.7.9
  • pinia 2.3.1

---

🔧 Detailed Setup

MLflow Configuration

The MLflow server needs to be running before starting the application:

mlflow server --host 127.0.0.1 --port 4800

Configuration in config.py:

MLFLOW_URI = "http://127.0.0.1:4800"
EXPERT_SCORE_CHAMPION = "models:/ChampionModel@expert_score_champion"
WEAKEST_LINK_CHAMPION = "models:/ChampionModel@weakest_link_champion"
POSE_TRANSFORMER_URI = "models:/PoseCoordinateTransformer@coordinate_transformer_champion"
DEPTH_ESTIMATOR_URI = "models:/PoseDepthEstimator@depth_estimator_champion"
SEQUENCE_DETECTOR_URI = "models:/SquatSequenceDetector@sequence_detector_champion"
FORM_CLASSIFIER_URI = "models:/SquatFormClassifier@form_classifier_champion"
PERFORMANCE_SCORER_URI = "models:/SquatPerformanceScorer@performance_scorer_champion"

Backend Setup

1. Create and activate virtual environment:

   cd backend
   python -m venv venv
   source venv/bin/activate  # Windows: venv\Scripts\activate

2. Install dependencies:

   pip install -r requirements.txt

3. Start the FastAPI server:

   uvicorn main:app --host 0.0.0.0 --port 5000

Frontend Setup

1. Install dependencies:

   cd frontend
   npm install

2. Start development server:

   npm run dev  # Development mode
   # or
   npm run build && npm run preview  # Production mode

Project Structure

Project_in_data_intensive/
├── backend/                 # FastAPI backend application
│   ├── src/                # Source code
│   │   ├── app.py         # Main application entry point
│   │   ├── config.py      # Configuration and model URIs
│   │   ├── controllers/   # API route controllers
│   │   ├── services/      # Business logic services
│   │   ├── schemas/       # Pydantic data models
│   │   ├── utils/         # Utility functions and scalers
│   │   └── tests/         # Unit and integration tests
│   ├── requirements.txt   # Python dependencies
│   └── pyproject.toml    # Project configuration
├── frontend/               # Vue.js frontend application
│   ├── src/               # Source code
│   │   ├── components/    # Vue components
│   │   ├── views/         # Page views
│   │   ├── services/      # API services
│   │   └── router/        # Application routing
│   ├── public/            # Static assets and 3D models
│   └── package.json       # Node.js dependencies
├── ML/                     # Machine learning experiments
│   ├── classification/    # Classification models notebooks
│   ├── linearregression/  # Regression models notebooks
│   └── deeplearning/      # Deep learning models notebooks
├── reports/                # Assignment reports
└── training_data/          # Pose data storage

Jupyter Notebook Integration

Set MLflow tracking URI in your notebooks:

import mlflow
mlflow.set_tracking_uri("http://127.0.0.1:4800")

---

🔍 Verification

1. MLflow UI should be accessible at http://127.0.0.1:4800
2. Backend API should be accessible at http://localhost:5000
3. Frontend should be accessible at http://localhost:4173 (production) or http://localhost:5173 (development)

API Endpoints

The backend provides the following main endpoints:

• POST /api/v1/expert-score/predict - Expert score prediction
• POST /api/v1/weakest-link/predict - Weakest link classification
• POST /api/v1/pose/upload - Pose data upload and analysis
• GET /api/v1/pose/sessions - List pose analysis sessions
• GET /api/v1/pose/sessions/{session_id}/feedback - Get comprehensive feedback

---

🏗️ Architecture Overview

Backend Services Architecture

The backend follows a service-oriented architecture with the following key services:

1. Core ML Services

   • LinearRegressionService: Expert score predictions
   • ClassificationService: Weakest link classification

2. Pose Analysis Pipeline

   • KeypointStorageService: Pose data persistence
   • FeatureEngineeringService: Biomechanical feature extraction
   • ExerciseDetectionService: Squat sequence identification
   • Pose2Dto3DService: 2D to 3D coordinate transformation
   • SquatFormClassificationService: Form quality assessment
   • SquatPerformanceScoringService: Performance quantification
   • SquatFeedbackService: Comprehensive feedback generation
   • PoseOrchestratorService: Central coordination hub

3. Controllers

   • RegressionController: Handles regression prediction requests
   • ClassificationController: Manages classification requests
   • PoseController: Orchestrates pose analysis workflows

---

🛠 Troubleshooting

Common Issues

1. Port Conflicts

   • Ensure ports 4800, 5000, and 4173 are available.
   • Check for other services using these ports.

2. MLflow Connection Issues

   • Verify the MLflow server is running.
   • Check URI configuration in config.py.
   • Ensure consistent URIs across all components.

3. Backend Issues

   • Verify all dependencies are installed.
   • Check Python version compatibility.
   • Ensure the virtual environment is activated.

4. Frontend Issues

   • Clear npm cache if needed:

     npm cache clean --force
     Remove-Item -Recurse -Force node_modules
     npm install

   • Check URI configuration in config.js.

5. Model Loading Issues

   • Verify all required models are registered in MLflow
   • Check model aliases are correctly set
   • Ensure scaler files exist in backend/src/utils/ directory
   • Make sure text and text are present during dev with the required models

6. Pose Detection Issues

   • Verify TensorFlow.js models are properly loaded
   • Check camera permissions in browser
   • Ensure sufficient lighting for pose detection

---

📝 Development Notes

• MLflow Server: Must remain running during development.
• Backend Dependency: Ensure the virtual environment is activated before running commands.
• Frontend: If issues arise, clear the npm cache or reinstall dependencies.
• For additional details on coding standards, linters and development workflow, please check our GitLab Wiki:
  Design-and-Coding-Rules-Summary

---

🤝 Contributing

For detailed guidelines on commit message conventions, branch naming, and versioning, please refer to our GitLab Wiki:
Design-and-Coding-Rules-Summary

---