Create an experiment
Creates an experiment and its first run, a draft named Run 1. When dataset_id is set, the run gets one row per datapoint, up to 1000. project_id defaults to the caller’s project and is required for API keys that span several projects.
Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Body
Experiment name, 1 to 255 characters.
Project the experiment is created in. Defaults to the caller's project: the API key's project, or the project open in the AI Studio. Required for API keys that span several projects.
Dataset whose datapoints become the rows of the first run, at most 1000. Must be in the experiment's project. Omit to start with an empty run.
Run to fork. Its task, metric, evaluator and review columns are copied
into the new experiment's first run; rows still come from dataset_id.
Must be in the same project as the new experiment.
Response
OK
An experiment: a dataset evaluated by one or more runs.
Experiment ID (ULID).
Experiment name.
Project that owns the experiment.
Dataset whose datapoints become the rows of each new run. Empty when the experiment was created without a dataset.
Run shown by default in the AI Studio, and the run whose columns Create a run copies. Create a run selects the new run; deleting the selected run selects the newest remaining one.
Creation time.
Time of the last update to the experiment.
Origin of the experiment: evaluatorq for results uploaded by the
evaluatorq framework; empty or manual otherwise.
Row height in the AI Studio grid: 1 (compact), 2 or 3 (tallest). Set to 1 on create.
Evaluators used by the experiment's evaluator columns, one entry per evaluator.
ID of the user or API key that created the experiment.