Introduction

MLflow is an open-source platform for managing the end-to-end machine learning lifecycle. Developed by Databricks, MLflow helps data scientists and ML engineers track experiments, package code, manage models, and deploy to production.

• Reproducibility: Track all experiments in detail
• Collaboration: Share results with team
• Model versioning: Manage different model versions
• Deployment ready: Easily deploy models to various platforms
• Framework agnostic: Works with TensorFlow, PyTorch, Scikit-learn, etc.

MLflow Main Components

MLflow consists of 4 main components:

Installation and Setup

Basic Installation

# Install MLflow
pip install mlflow

Install with extras for various backends
pip install mlflow[extras]

Verify installation
mlflow --version

Setup Database Backend (PostgreSQL)

For production, use a database backend:

# Install dependencies
pip install psycopg2-binary

Setup PostgreSQL (example using Docker)
docker run -d \
  --name mlflow-db \
  -e POSTGRESUSER=mlflow \

  -e POSTGRESPASSWORD=mlflow \
  -e POSTGRESDB=mlflow \

  -p 5432:5432 \
  postgres:13

# Install boto3 for S3 compatibility
pip install boto3

Setup MinIO (S3-compatible storage)
docker run -d \
  --name mlflow-minio \
  -p 9000:9000 \
  -p 9001:9001 \
  -e MINIOROOTUSER=minioadmin \

  -e MINIOROOTPASSWORD=minioadmin \

  minio/minio server /data --console-address ":9001"

# Development mode (local file store)
mlflow server --host 0.0.0.0 --port 5000

Production mode (with database and S3)
mlflow server \
  --backend-store-uri postgresql://mlflow:mlflow@localhost:5432/mlflow \
  --default-artifact-root s3://mlflow-artifacts \
  --host 0.0.0.0 \
  --port 5000

# MLflow Tracking
MLFLOWTRACKINGURI=http://localhost:5000


S3/MinIO Configuration
AWSACCESSKEYID=minioadmin
AWSSECRETACCESSKEY=minioadmin

MLFLOWS3ENDPOINTURL=http://localhost:9000

MLflow Tracking: Experiment Tracking

Basic Tracking

import mlflow
import mlflow.sklearn
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import loadiris

from sklearn.modelselection import traintestsplit
from sklearn.metrics import accuracyscore, f1score

Set tracking URI
mlflow.settrackinguri("http://localhost:5000")

Set experiment
mlflow.setexperiment("iris-classification")


Load data
iris = loadiris()
Xtrain, Xtest, ytrain, ytest = traintestsplit(
    iris.data, iris.target, testsize=0.2, randomstate=42
)

Start MLflow run
with mlflow.startrun(runname="random-forest-v1") as run:
    # Log parameters
    params = {
        "nestimators": 100,

        "maxdepth": 5,
        "randomstate": 42

    }
    mlflow.logparams(params)

    # Train model
    model = RandomForestClassifier(params)

    model.fit(Xtrain, ytrain)

    # Make predictions
    ypred = model.predict(Xtest)

    # Log metrics
    metrics = {
        "accuracy": accuracyscore(ytest, ypred),

        "f1score": f1score(ytest, ypred, average="weighted")

    }
    mlflow.logmetrics(metrics)

    # Log model
    mlflow.sklearn.logmodel(

        model, 
        "model",
        registeredmodelname="iris-classifier"

    )