Instalar MLflow y empezar a registrar

You run dozens of experiments and forget which settings gave the best score. MLflow records every run so no result is ever lost again.

MLflow is just a Python package. One pip install gives you the library, a tracking API, and a built-in web UI to browse runs.

The tracking server is where runs are stored. Start one locally and it serves an API plus a dashboard on port 5000.

Tell MLflow where the server lives with set_tracking_uri. After this, every run you log flows to that server.

A run is one execution of your training code. It bundles the parameters you chose, the metrics you got, and any files you saved.

Wrap your training in start_run. MLflow opens a run, gives it a unique ID, and closes it cleanly when the block ends.

An experiment is a named folder for related runs. Set one so all your churn-model attempts live together instead of scattered.

Behind the scenes MLflow writes run metadata to a backend store, like a SQLite or Postgres database, so it survives restarts.

No server yet? MLflow falls back to a local mlruns folder. Great for quick tests, but a shared server is better for teams.

Browse everything you logged in the MLflow UI. Just visit the server URL in a browser to see your runs as a sortable table.

Tracking is the foundation of MLOps. Once every run is captured, you can compare, reproduce, and promote models with confidence.

Let us check how you connect your code to MLflow.

You installed MLflow, started a tracking server, pointed your code at it, and opened a run. Every experiment now has a home. ✅