How to compare the explainability scores of different samples

Selecting the samples to be compared

To select the data points you are interested in comparing, first, go to the report page, inside the corresponding project.


Don't have a project yet?

The explainability scores live inside a project. If you haven’t created a project yet, make sure you create it first. In case you missed it, here is a tutorial about it.

On the report page, you will notice that below the error analysis panel, a data slice is shown.

Using the checkbox, click on the data rows that you would like to compare. It can be as few as two data points. After selecting the data points of interest, a Compare button should appear on the filter bar.

Click on the Compare button to compare the selected data points side-by-side.

Comparing samples side-by-side

With this view, you can compare the explainability scores and the model’s predictions any two of the selected points.

You can toggle between selected rows to compare using the data table at the bottom of the page. The top two rows in the data table represent the two data points currently being compared: the base data point in the panel is highlighted in blue and the comparison point lies under it.

To compare to another selected point, click on a row from the second group of rows in the data table. If there are none, return to the previous page and select more than two rows to compare.


Interested in NLP?

The image examples we displayed use tabular data. The exact same steps apply for NLP, if that is the type of task you are interested in.

Did this page help you?