Skip to main content
Version: v1.0.0

Prediction intervals support

To enable support for prediction intervals in H2O MLOps, set requestPredictionIntervals parameter to true. Note that if prediction intervals are not supported by the model or not returned for some reason, H2O MLOps will either leave the field empty or return an error response. This mechanism ensures that you are aware when prediction intervals are not available.

Once enabled, the prediction intervals are returned as an array. For each prediction, a lower and upper bound are returned.

note

Prediction interval support is currently only available for regression models in all MOJO runtimes and the Driverless AI scoring pipeline runtime.

Step 1: Check if the deployment has requestPredictionIntervals support in a curl request​

Use the /capabilities endpoint to confirm that the model supports prediction intervals.

curl -X GET https://<DEPLOYMENT_URL>/model/capabilities

["SCORE_PREDICTION_INTERVAL","SCORE","CONTRIBUTION_ORIGINAL","CONTRIBUTION_TRANSFORMED"]

The SCORE_PREDICTION_INTERVAL capability indicates that prediction intervals are supported.

To try this using H2O MLOps Python client, see View scorer capabilities.

Step 2: Make a prediction with requestPredictionIntervals enabled​

To request prediction intervals using the H2O MLOps Python client, see Prediction intervals.

Sample output:

{'fields': ['score'],
 'id': '1b0488b6-ee91-11ed-a05d-4ab989c17db4',
 'predictionIntervals': {'fields': ['score.lower',
   'score.upper'],
  'rows': [['55524.044704861124', '371673.57907986105']]},
 'score': [['200005.90798611112']]}
Feedback