iris-mlops

An inference server for iris classification using Flask with MLflow integration, containerized with Docker, designed for ML lifecycle management and deployment.

---

Features

• Train, evaluate, register, and stage ML models with MLflow
• REST API for inference with live model reloading
• Dockerized app and MLflow tracking server
• Metrics endpoint with latency, status code, and endpoint statistics
• Periodic background model reload for zero downtime

---

Prerequisites

• Python 3.11+ (for local dev)
• Docker & Docker Compose (for containerized deployment)
• Git (optional, for tracking SHA in MLflow)

---

Setup Locally

1. Clone the repo:

   git clone https://github.com/ivarunseth/iris-mlops.git
   cd iris-mlops

2. Create and activate virtual environment:

   python3.11 -m venv venv
   source venv/bin/activate

3. Install dependencies:

   pip install -U pip setuptools wheel
   pip install -r requirements.txt

4. Setup environment variables:

   Create .env file (optional):

   MLFLOW_TRACKING_URI=http://localhost:5000
   MLFLOW_MODEL_NAME=iris_classifier
   

---

Setup with Docker

1. Build the Docker image:

   docker compose build

2. Start containers:

   docker compose up

3. The MLflow UI is available at http://localhost:5000
   API server available at http://localhost:5001/api/health

---

Running Locally

Start MLflow Tracking Server

mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root ./mlflow/mlruns --host 0.0.0.0 --port 5000

Run Flask API

python app.py

---

Running with Docker

Start all services with Docker Compose:

docker compose up

---

Training, Evaluating, Registering & Staging Models

Locally

Running in host system locally

python -m src.train.py
python -m src.register.py
python -m src.stage.py <promote/demote>
python -m src.evaluate.py <None/staging/production>

With Docker

Run training inside the app container:

docker exec -it iris-mlops-app-1 python -m app.src/train.py
docker exec -it iris-mlops-app-1 python -m app.src/register.py
docker exec -it iris-mlops-app-1 python -m app.src/stage.py <promote/demote>
docker exec -it iris-mlops-app-1 python -m app.src/evaluate.py <None/staging/production>

This script will:

• Train models
• Register the best model to MLflow Model Registry
• Stage the best model (to staging or production depending on config)
• Evaluate models on given stage name (alias)

---

Model Inference

Locally

Send a POST request to predict endpoint:

curl -X POST http://localhost:5001/api/predict \
-H "Content-Type: application/json" \
-d '{"inputs": [{"sepal length (cm)": 5.0, "sepal width (cm)": 1.0, "petal length (cm)": 4.0, "petal width (cm)": 1.0}]}'

With Docker

Same as local but using the Docker mapped port.

---

Troubleshooting & Notes

• Permission errors with MLflow artifacts: Ensure volume directories have correct ownership matching container UID/GID.
• Git warnings in MLflow: Install git or set GIT_PYTHON_REFRESH=quiet environment variable.
• Model not loaded: API auto-reloads models periodically; manual reload can be implemented if needed.