MLflow Experiment Tracking

Overview

Tracking ML experiments, parameters and metrics with MLflow. This guide covers practical implementation for production ML systems.

Why This Matters in MLOps

Setup

bash
Install required tools
pip install mlflow mlflow pandas numpy scikit-learn
Or with Docker
docker pull ghcr.io/mlflow/mlflow

Core Implementation

python
import os
import json
import logging
from datetime import datetime
from pathlib import Path
logger = logging.getLogger(__name__)
class MLflowExperimentTracking:
    """
    MLflow Experiment Tracking implementation.
    
    Handles: experiment management
    Tool: mlflow
    """
    
    def __init__(self, config: dict = None):
        self.config = config or self._default_config()
        self._setup()
    
    def _default_config(self) -> dict:
        return {
            "tool": "mlflow",
            "environment": os.getenv("ENVIRONMENT", "development"),
            "log_level": "INFO",
        }
    
    def _setup(self):
        """Initialize mlflow connection and resources."""
        logging.basicConfig(level=self.config.get("log_level", "INFO"))
        logger.info(f"Initialized MLflow Experiment Tracking with config: {self.config}")
    
    def run(self, **kwargs) -> dict:
        """Execute experiment management."""
        start = datetime.utcnow()
        
        try:
            result = self._execute(**kwargs)
            
            elapsed = (datetime.utcnow() - start).total_seconds()
            logger.info(f"MLflow Experiment Tracking completed in {elapsed:.2f}s")
            
            return {
                "status": "success",
                "result": result,
                "elapsed_seconds": elapsed
            }
        
        except Exception as e:
            logger.error(f"MLflow Experiment Tracking failed: {e}")
            return {
                "status": "failed", 
                "error": str(e)
            }
    
    def _execute(self, **kwargs) -> dict:
        """Core experiment management logic. Override to customize."""
        return {"completed": True, "tool": "mlflow"}
Configuration
config = {
    "tool": "mlflow",
    "tracking_uri": os.getenv("MLFLOW_TRACKING_URI", "http://localhost:5000"),
    "artifact_root": "./artifacts",
}
Initialize
processor = MLflowExperimentTracking(config)
result = processor.run()
print(json.dumps(result, indent=2))

MLFLOW Integration

python
Specific mlflow integration for experiment management
import subprocess
def setup_mlflow():
    """Configure mlflow for experiment management."""
    
    # Initialize project
    print(f"Setting up mlflow for experiment management...")
    
    # Example configuration
    config = {
        "project": "my-ml-project",
        "tool": "mlflow",
        "specialty": "experiment management",
        "version": "1.0.0"
    }
    
    # Save configuration
    Path(".mlflow").mkdir(exist_ok=True)
    with open(f".mlflow/config.json", "w") as f:
        json.dump(config, f, indent=2)
    
    print(f"mlflow configured for experiment management")
    return config
config = setup_mlflow()

Monitoring and Alerting

python
from dataclasses import dataclass
import time
class MLOpsMonitor:
    """Monitor experiment management metrics."""
    
    def __init__(self):
        self.metrics: list[MetricSnapshot] = []
        self.thresholds = {
            "error_rate": 0.05,
            "latency_p99_ms": 1000,
            "data_drift_score": 0.3
        }
    
    def record(self, metric: str, value: float, labels: dict = None):
        snapshot = MetricSnapshot(
            timestamp=time.time(),
            metric_name=metric,
            value=value,
            labels=labels or {}
        )
        self.metrics.append(snapshot)
        self._check_threshold(metric, value)
    
    def _check_threshold(self, metric: str, value: float):
        threshold = self.thresholds.get(metric)
        if threshold and value > threshold:
            logger.warning(f"ALERT: {metric}={value:.3f} exceeds threshold {threshold}")
monitor = MLOpsMonitor()

CI/CD Integration

yaml
.github/workflows/ml-pipeline.yml
name: ML Pipeline
on:
  push:
    paths: ['src/', 'data/']
jobs:
  train-and-evaluate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      
      - name: Setup Python
        uses: actions/setup-python@v5
        with:
          python-version: '3.11'
      
      - name: Install dependencies
        run: pip install -r requirements.txt
      
      - name: Run experiment management
        run: python -m src.mlflow_experiment_tracking
        env:
          MLFLOW_TRACKING_URI: ${{ secrets.MLFLOW_URI }}
      
      - name: Check model quality
        run: python -m src.validate_model

Best Practices

Resources