Experiments

An experiment is a saved training setup: compute, environment or dataset, algorithm, training settings, and optional hyperparameter search. While status is Draft, you edit it in the wizard. After you Train, status moves through Pending, Running, and finished states.

Finding experiments

Start from Experiments in the sidebar.

What you want

Where in the UI

List projects

Experiments sidebar (project list)

Project workspace

Click a project

New experiment (wizard)

Project → ExperimentsNew experiment

Open wizard

Click the experiment name → Edit experiment (draft) or View experiment (non-draft)

Training logs

Row menu () → View logs

More screen names: Finding your way.

Project workspace tabs

Inside a project you will see Experiments (the main table with status, Train, Resume, row menus, and checkpoints), Results (compare runs and open logs), and Pipelines on Advanced Training projects only.

Draft vs running

Only Draft experiments are fully editable. After you schedule training, you can still walk the wizard steps to read settings, but fields are not saved the same way as in a draft.

The experiments table shows Train when the row is runnable (valid algorithm and environment or dataset, and status Draft, Succeeded, or Stopped). Resume appears for finished or stopped runs that have checkpoints, not while status is Draft, Running, or Pending.

Status names: Experiment statuses.