Explainability

Explain predictions to enhance model transparency, ensure ethical and unbiased outcomes, and gain insights into the underlying data patterns.

Deploy an explainer

There are multiple different ways to deploy an explainer when creating a Deployment. For details information, see Deploying an explainer.

Explain a prediction

We assume that you have deployed a model with an explainer. Use the UI, Deeploy API, or Python client to explain a prediction.

UI

Navigate to the Test page in your Deployment

Add input data. It's only possible to explain a single input at a time on the **Test** page. Select Explain prediction and click Make prediction.

The explanation result is part of the response. If available, a visualization of the explanation is shown below the response.

Deeploy API

Consult the inference section within the Swagger documentation

Python

Consult the predictions and explanations section within the Python client documentation

Explainability visualizations

Deeploy provides a visualization of explanation output for the standard explainers and some integrated explainers. Visualizations are currently available for:

Tree SHAP

Saliency

Attention

Captum text explainer dashboard

Captum image explainer dashboard

For details on how to deploy one of these explainers, refer to Deploying an explainer.

To view the visualization for an explained prediction, navigate to that prediction on the Predictions page and click View visual explanation.