Issues in formulas for K-Nearest Neighbors, Naive Bayes and SVM. JMP Pro 15

Question

Dear colleagues,

As a user of JMP Pro 15, I might understand it wrong and I would highly appreciate your assistance. It may be just a misunderstanding in JMP's functionality.

Well, in the formula in-built into the column after saving from KNN there is a red line indicating some issues in the equation's structure. And one more thing: while generating a code, e.g. in Python in the menu of publishing formula, e strange report appears. For both cases I have attached the snapshots.

Similarly, generating codes for SVM or Naive Bayes brings this issue.

I am slow in programming and, please, clarify these unknown messages. I am a chemist who utilizes JMP for scientific cases.

For your help I will be very thankful.

Michael

Mark_Bailey · Accepted Answer

These look like software bugs with the score code generator. I would contact JMP Technical Support (support@jmp.com) about this matter.

nascif_jmp · Accepted Answer

Please notice that JMP is providing an alert explaining what the problem is: the operator has not been implemented yet. This is a side effect our current implementation. Given the breadth of the JSL language, and the number of deployment languages we support, we implemented the operators used by the most commonly used models in our first version of this feature, and over time - and based on customer input - we might be adding more. So yes, please contact Technical Support so that we can track the demand for the missing operators. Better yet, add a request to the JMP Wish List.

This is an unfortunate limitation of our implementation, but we thought it would have the benefit of providing our customers with as much functionality as possible.

For scenarios where the model is not fully supported (uses operators we don't cover yet) the current design allows customers to keep working by giving them the opportunity to provide their own implementation for the missing operators. In the generated Python code you will see that it makes a call to 'placeholder' functions for the unsupported operators.

if _temp_1:
  _temp_3 = np.nan
else:
  _temp_3 = 2.25012937537175 + placeholder_Normal_Log_Density((-2.33428499519846 + 
9.48896339511568 * indata[u"Petal width"]))

If a definition for those functions that matches the JSL behavior is provided, the model will be fully functional.

Nazarkovsky · Accepted Answer

Dear Nascif,

Many thanks for such a thorough response. As our colleague, Mark Bailey, and you also suggested, I'd better inform Technical Support and put this claim into the Wish list to enhance JMP's functionality. I liked the idea.

Wish you a nice weekend!

Regards,

Michael