30 Days of MLOps Challenge · Day 7

MLflow iconModel Experiment Tracking with MLflow – Log It or Lose It

By Aviraj Kawade · June 26, 2025 · 5 min read

Track parameters, metrics, and artifacts across runs so you can reproduce, compare, and collaborate effectively.

← Previous: Day 6 Next: Day 8 → Back to Roadmap
💡 Hey — It's Aviraj Kawade 👋

Key Learnings

  • Why experiment tracking is critical (reproducibility, comparability, collaboration).
  • MLflow components: Tracking, Projects, Models, Registry.
  • Track params/metrics/artifacts with the Tracking API.
  • Log models and register for deployment.

What is ML Experiment Tracking?

Logging, organizing, and comparing ML runs to know which changes improved results and to reproduce them reliably.

ML experiment tracking overview

What is MLflow?

Open‑source platform for managing the ML lifecycle: Tracking, Projects, Models, and Registry. Works with any ML library.

ComponentPurpose
TrackingLogs and queries experiments
ProjectsReproducible runs with environment definition
ModelsStandard model packaging across frameworks
RegistryCentral store with versions and stages

Track with the MLflow API

pip install mlflow

import mlflow

mlflow.set_experiment("my-experiment")

with mlflow.start_run(run_name="my-first-run") as run:
    mlflow.log_param("learning_rate", 0.01)
    mlflow.log_param("optimizer", "adam")
    mlflow.log_metric("accuracy", 0.95)
    mlflow.log_metric("loss", 0.05)
    with open("model.txt", "w") as f:
        f.write("This is a dummy model file.")
    mlflow.log_artifact("model.txt")
    print("Run ID:", run.info.run_id)

for step in range(5):
    mlflow.log_metric("accuracy", 0.9 + step*0.01, step=step)

Accessing Run Data

from mlflow.tracking import MlflowClient

client = MlflowClient()
runs = client.search_runs(experiment_ids=["0"])  # default experiment id
for run in runs:
    print("Run ID:", run.info.run_id)
    print("Params:", run.data.params)
    print("Metrics:", run.data.metrics)

UI Access

# Local UI
mlflow ui
# Or as a server
mlflow server --host 127.0.0.1 --port 8080 \
  --backend-store-uri sqlite:///mlruns.db \
  --default-artifact-root ./artifacts
MLflow UI screenshot

Log, Register, and Deploy

import mlflow, mlflow.sklearn
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y)
model = RandomForestClassifier().fit(X_train, y_train)

with mlflow.start_run() as run:
    mlflow.sklearn.log_model(model, artifact_path="rf_model", registered_model_name="IrisRFModel")

from mlflow.tracking import MlflowClient

client = MlflowClient()
client.create_registered_model("IrisRFModel")
client.create_model_version(
    name="IrisRFModel",
    source=f"{run.info.artifact_uri}/rf_model",
    run_id=run.info.run_id
)

# Load from Registry
model_uri = "models:/IrisRFModel/Production"
loaded_model = mlflow.sklearn.load_model(model_uri)
# Serve
# mlflow models serve -m models:/IrisRFModel/Production -p 5001

Challenges

  • Add MLflow tracking to a Sklearn or TensorFlow script.
  • Log at least 3 params and 2 metrics.
  • Save your trained model as an artifact.
  • Run the UI and compare at least 2 runs.
  • Use a file backend and custom artifact root.
  • Try mlflow.projects with conda.yaml or Dockerfile.
← Back to MLOps Roadmap