Skip to main content

Try Out View

The try out view is used for evaluating a model before deploying it and integrating it into external systems. It can be accessed by opening a model from the Model Management view and clicking on the Try it out button in the navigation section on the right side.

note

Only models that are ready and active can be tried out — the Try it out button is disabled otherwise.

Here you can send requests to the model either by filling in the fields or by editing the raw JSON (toggle Raw JSON at the top of the Variables column).

Predictions and other model outputs are shown on the right in the Prediction column.

try out

Explanations​

Models that are configured with explanations: default (see Edit the metadata) and have their training data assigned receive SHAP explanations that assist the user in determining which features had the most impact on the outcome of the prediction in comparison to other features.

Taking the above example, you can see that Melting Pressure had the biggest impact on the Weight being 88.419... whereas Torque negatively impacted the outcome.

predictions

Users can also configure their own explanations giving them flexibility to visualize the data according to their use case.

tip

See the Custom Explanations guide for more details regarding explanations.

Out-of-Distribution Detection​

If the model has OOD detection enabled (oodDetection: default in the model metadata) and training data assigned, the try out view also indicates whether the request you sent is out-of-distribution, i.e. how much it differs from the data the model was trained on. This helps you judge how trustworthy a prediction is. The same detection is applied to requests of deployed models and is visible on the Monitoring view.