Analyze runs
In the Neptune app, you can explore individual runs, or visualize multiple runs to compare their metadata.
Show experiment head runs or all runs
By default, Neptune only shows head runs of experiments.
To list all runs in the table view, as opposed to only the latest run of each experiment, switch from Experiments to Runs in the toolbar:
Select which runs are visible
To plot runs on charts, in the left table pane, click the eye icons ().
To access display options, click the eye icon () on the header row.
You can plot a maximum of 1000 graphs on a single chart. For example, if you're displaying 20 metrics from each run, you can select a maximum of 50 runs to compare.
Exploring a single run
Click on a run to view all of its metadata. You can visualize it in the following tabs:
Tab | Keyboard shortcut | Description |
---|---|---|
Attributes | Navigate the complete metadata structure and preview attributes. | |
Charts | See metrics (such as accuracy or loss) visualized as graphs. | |
Dashboards | Compose your own set of custom widgets. |
To show the most relevant metadata in a single view, create a custom dashboard and then open runs one by one to compare their data.
To get a link to the run in the Neptune web app:
from neptune_scale import Run
run = Run(...)
run.get_run_url()
For details, see Construct Neptune URLs.