Dashboards

The dashboard view. Marked are the dashboard name, add new dashlet panel, and the tabs (insights/issues/tests)

The Dahboard view is the view in which all Model analysis is made within the platform. This view supports multiple dashlets: population exploration, metrics related, sample analysis - each of these enables a separate analysis flow.

In addition, it offers a suite of capabilities:

  • The insights panel - that pinoints to sepcific model & data failures

  • The issues panel - that supports an in-platform log of detected failures, and enables team collaboration around the solution of such failures.

  • The tests panel - that support an in-platform tracking of important failure cases and allows a one-click overview to test for model regressions.

Adding and removing a model from analysis

Adding and removing models from analysis

To add or remove models from the analysis, we use the versions tab. Next to each model run version that was previously evaluated within tensorleap, there's an "Add To Dashboard" button. Clicking the button adds the model to the dashboard, populating all existing dashlets, insights, and tests using the added model.

The currently examined model runs could be seen in color-coded cards at the top of the dashboards. Each dashlet would be split according to the number of models concurrently examined.

To remove a model that was added to the analysis, either click the "x" button next to its card at the top, or click the corrosponding "Add To Dashboard" icon again to deselect it.

Adding Dashlets

Adding a new dashlet in Tensorleap

Adding a dashlet is done from the "Add New Dashlet" dropdown in the top right. Available dashlets for adding are:

Creating multiple dashboards

Dashboard selection panel

A project can hold more than one dashboard. To add a dashboard, click the "+" icon in the top left othe dashboard. To switch between dashboards, click the dropdown panel in the top left side of the dashboard.

Last updated

Was this helpful?