Skip to content
BENIBIX Learn

Experiment Tracking

Also known as ML Experiment Tracking, Run Tracking

Notes on each training run. You can then compare them and redo the best one.

Think of it like

Think of it like a science notebook. You write down what you tried and what happened, every time.

Example

Emma tries 40 versions of a model in a week. Her tracking tool shows that run 27 scored best, with a smaller learning rate, so she knows how to rebuild it.

Why it matters

It turns guesswork into a record. You can always say how a model was made and why it was picked.

Definition

Experiment tracking means saving a record of every training run, so you can compare runs and repeat the best one. Each record holds the settings used, the version of the code and the data, the scores and the files the run made. A dashboard then shows runs side by side, so you can see which change helped. Without it, people forget which settings gave the best model. It is often the first MLOps tool a team adds.
Each run is a row: settings, data version and score. The best one is marked.

Common mistake

Tracking scores but not the data version. If the data changed between runs, the scores can't be fairly compared.

Key takeaways

  1. 1
    Every run saved. Settings and scores.
  2. 2
    Code and data versions. Both recorded.
  3. 3
    Compare side by side. Pick the best.

Compare