MLOps Project

This repository contains a complete MLOps project for predicting customer subscription likelihood using machine learning models. The project includes model training, MLflow tracking, API service for model serving, and containerization.

Project Structure

• api_service.py: FastAPI service for model predictions with caching and error handling
• main.ipynb: Main Jupyter notebook containing data analysis and model development
• train_and_log_lr.py: Script for training and logging the LogisticRegression model
• mlflow_log_model.py: Utilities for logging models to MLflow
• preprocessing.py: Data preprocessing utilities
• test_api.py: API service tests
• test_customer_profiles.py: Test cases for different customer profiles
• Dockerfile: Container definition for the API service
• docker-compose.yml: Service orchestration configuration
• data/: Directory containing dataset files
• models/: Saved model files
• mlruns/: MLflow tracking information

Features

• Machine Learning Models:
  • Random Forest Classifier
  • Logistic Regression
• MLflow Integration:
  • Model versioning
  • Experiment tracking
  • Metrics logging
• FastAPI Service:
  • Model prediction endpoints
  • Request validation
  • Error handling
  • Response caching
  • CORS and Gzip middleware
• Docker Support:
  • Containerized API service
  • Multi-stage builds
  • Environment configuration

Setup

1. Clone the repository:

   git clone https://github.com/Jsznn/mlops-project-2.git
   cd mlops-project-2

2. Install dependencies:

   pip install -r requirements.txt

3. Run the API service:

   python api_service.py

   Or using Docker:

   docker-compose up

API Usage

The API provides endpoints for making predictions using both Random Forest and Logistic Regression models:

Endpoints

• GET /: Root endpoint with API information
• GET /models: List available models and their versions
• POST /predict/best_rf: Get predictions from Random Forest model
• POST /predict/best_lr: Get predictions from Logistic Regression model

Example Request

curl -X POST "http://localhost:8000/predict/best_rf" \
  -H "Content-Type: application/json" \
  -d '{
    "usia": 24,
    "pekerjaan": "manajer",
    "status_perkawinan": "lajang",
    "pendidikan": "Pendidikan Tinggi",
    "gagal_bayar_sebelumnya": "no",
    "pinjaman_rumah": "yes",
    "pinjaman_pribadi": "no",
    "jenis_kontak": "cellular",
    "bulan_kontak_terakhir": "jul",
    "hari_kontak_terakhir": "fri",
    "jumlah_kontak_kampanye_ini": 2,
    "hari_sejak_kontak_sebelumnya": 999,
    "jumlah_kontak_sebelumnya": 0,
    "hasil_kampanye_sebelumnya": "nonexistent",
    "tingkat_variasi_pekerjaan": -1.7,
    "indeks_harga_konsumen": 94.215,
    "indeks_kepercayaan_konsumen": -40.3,
    "suku_bunga_euribor_3bln": 0.885,
    "jumlah_pekerja": 4991.6,
    "pulau": "Papua"
  }'

Example Response

{
  "model_name": "best_rf",
  "prediction": 1,
  "probability_subscribe": 0.7555,
  "model_version": "local"
}

Development

• Run tests: python -m pytest test_api.py test_customer_profiles.py
• Check API documentation: Visit http://localhost:8000/docs when API is running
• Monitor MLflow: Start MLflow UI with mlflow ui

License

MIT