Snowflake UDF Jar Prediction with Explanation

[xwang]

Hi,

I have a question about how to retrieve more than 3 explanations when I deploy the .jar model on Snowflake as a UDF.

What I have done is to call this handler in the udf

handler = 'com.datarobot.prediction.simple.ClassificationPredictor.scoreWithExplanations'

, however, by default, it only returned 3 explanations. I need 20 explanations.

Anybody can help? Thank you for your time and patience.

Best

Xu

[jbaekdahl]

Hi Xu,

scoreWithExplanations has a parameter that will let you customize the number of explanations which are returned.

You can create your UDF like this

CREATE OR REPLACE FUNCTION datarobot_deployment_628d25c5a05250cf6aed2f90(RowValue OBJECT, MaxCodes INTEGER, ThresholdLow FLOAT, ThresholdHigh FLOAT)
RETURNS OBJECT
LANGUAGE JAVA
IMPORTS=('@~/jars/628d25c5a05250cf6aed2f90-625e8e17eed2ffd50a04330d-pe.jar')
HANDLER='com.datarobot.prediction.simple.ClassificationPredictor.scoreWithExplanations';

and when scoring, specify 20 for the MaxCodes parameter:

SELECT datarobot_deployment_628d25c5a05250cf6aed2f90(OBJECT_CONSTRUCT_KEEP_NULL(*), 20, NULL, NULL)
FROM "input.csv"

You can find these code snippets in the Portable Predictions tab of a Snowflake deployment. Remember to enable "Include Prediction Explanations" to get the correct snippet.

See this page in the documentation for more info.