Skip to content

Experiments

The Experiments page is the entry point for OICM Tracking. It lists the experiments available in the selected workspace and lets users create, filter, open, and manage experiment metadata.

An experiment groups multiple runs that belong to the same tracking objective. For example, one experiment can contain two runs that train two different models on the same dataset.

Experiment Tracking

The page shows all active experiments in the workspace. Use search to find experiments by name.

Experiment tracking list

Creating Experiments

Experiments can be created in two ways.

Create from the UI

Use Create Experiment from the experiments page to create a new experiment. Enter the experiment name, add a short description, optionally add tags, and click Create experiment.

Create experiment dialog

Create from TrackingClient

Use set_experiment(...) to create or select an experiment before logging runs:

from oip_tracking_client.v2.tracking import TrackingClient

tc = TrackingClient(
    api_host="https://YOUR-OICM-HOST/api/tracking",
    api_key="YOUR_API_KEY",
)

tc.set_experiment(
    experiment_name="diabetes-model-comparison",
    workspace_id="YOUR_WORKSPACE_ID",
)

Use create_experiment(...) when tags should be added during experiment creation:

tc.create_experiment(
    experiment_name="diabetes-model-comparison",
    workspace_id="YOUR_WORKSPACE_ID",
    tags=["regression", "sklearn"],
)

You can also add experiment tags after selecting the experiment:

tc.set_experiment_tags(["example-experiments"])

Experiment Details

Click an experiment card to open the experiment details page. By default, the experiment opens on the Runs tab, which lists all runs created under that experiment.

A run is a single tracked execution inside an experiment. Runs are where TrackingClient logs the actual parameters, metrics, and tags generated by training or evaluation code. Multiple runs can exist under one experiment, which makes it easy to keep related attempts together.

You will learn more about runs, metrics, and creating runs in Run UI.

Experiment runs

Compare Runs

The Compare runs tab is used to compare metrics from two or more runs in the same experiment. This workflow is covered in more detail in Run UI after runs and metrics are introduced.

Activity Log

Use the Activity log tab to review experiment events in chronological order. For example, when runs are started under an experiment, the activity log shows the experiment name and run name for each event.

Experiment activity log

Settings

Use the Settings tab to update experiment metadata and manage the experiment lifecycle. From this page, you can edit the experiment description, add or remove tags, save changes, and delete the experiment when it is no longer needed.

Experiment settings

Experiment tags can be added with TrackingClient:

tc.set_experiment_tags(["regression", "baseline-study"])

Experiments can also be deleted through TrackingClient when you already have the experiment ID:

tc.delete_experiment("EXPERIMENT_ID")

Experiment descriptions are edited from the Settings tab in the UI.

Next Steps