You can request Prediction Explanations be returned with your predictions by setting the maxExplanations job parameter to a non-zero value. You can also set thresholds for computing explanations. If you do not configure a threshold, DataRobot computes explanations for every row.
Prediction Explanation Parameters
Job parameter
Description
Example value
maxExplanations
Optional. Compute up to this number of explanations.
10
thresholdHigh
Optional. Limit explanations to predictions above this threshold.
0.5
thresholdLow
Optional. Limit explanations to predictions below this threshold.
0.15
If Prediction Explanations are requested, DataRobot returns four extra columns for each explanation, in the format EXPLANATION_<n>_IDENTIFIER (where n is the feature explanation index, from 1 to the maximum number of explanations requested). The returned columns are:
EXPLANATION_<n>_FEATURE_NAME: The feature name this explanation covers.
EXPLANATION_<n>_STRENGTH: The feature strength as a float.
EXPLANATION_<n>_QUALITATIVE_STRENGTH: The feature strength as a string, a plus or minus indicator from +++ to ---.
EXPLANATION_<n>_ACTUAL_VALUE: The feature associated with this explanation.
If your deployment was configured with an association ID for accuracy, all result sets will have that column passed through from the source data automatically.
If your use case has a strict output schema that does not match the DataRobot output, you can rename and remove any columns from the output using the columnNamesRemapping job configuration property.
Output Column Remapping Parameters
Job parameter
Description
Example value
columnNamesRemapping
Optional. An object/dictionary containing a mapping of columns to rename from/to. Set a target column to null or false to ignore it; defaults to an empty set {}.