When you run a strategy that references a Predictive Model rule instance, an error that is related to the PMML file that is uploaded in the rule instance might occur.
An error occurs and a message is displayed when one of the following conditions applies:
- Missing or invalid input values for the model
- Data type mismatch for the model inputs
- Missing a value for a field in the model
- Invalid number of arguments in the model
- Error running a custom function in the model
- Unsupported or invalid feature in the model
- Problem with running the predictive model
For any type of error, ask a PMML developer to investigate the error message that indicates the nature of the error and fix the problem.
Solution - Missing or invalid input values for the model
- Check the strategy that references the predictive model and the PMML model that was uploaded in the predictive model.
- If values for the predictive model are missing, check the strategy for the model's inputs.
Run the predictive model and enter values for the inputs:
- Open the predictive model.
- Click .
- Enter values for the inputs, and click .
If the model runs successfully, the problem might be with the PMML file.
Ask the PMML developer who created the model to update the model definition. The missing values need to be handled by the missingValueReplacement attribute of the MiningField element.
If the error recurs, check whether the supplied values are valid. The correct values depend on the data type and possible values that are defined in the DataFieldelement of the PMML model definition.
If an error occurs after you provide the correct input values, there might be a problem with executing the predictive model.
Solution - Data type mismatch for the model inputs
The type of input data must match the type that is specified in the predictive model. You need to modify the predictive model or the strategy that references the predictive model. For example, the error occurs when a strategy provides a string value for the integer property in the predictive model.
Solution - Missing a value for a field in the model
Very often this problem occurs when you specify derived fields in the PMML model that get values from the parent field, and there is a problem with the parent field, for example, a derived Age field from the parent field
A PMML developer needs to investigate the derived and parent fields to provide the missing values for the model.
Solution - Invalid number of arguments in the model
The model definition uses functions and there is a discrepancy in arguments between the function definition (the DefineFunction element) and the function call (the Apply element). The PMML model might have been incorrectly generated or modified manually.
A PMML developer needs to investigate the function definition and its call.
Solution - Error running a custom function in the model
A PMML developer needs to debug the custom function that is defined in the Configurations tab of the predictive model. For more information, see Predictive Model rule form - Completing the Configurations tab.
Solution - Unsupported or invalid feature in the model
Modify the PMML model. For more information, see Supported models for import.
Check the model definition for valid input. If the problem is more complex and caused by issues with the PMML model, you might need to ask a system architect and PMML developer for help.