The period (.) in <output> creates problems

[AayushSameerShah]

Classification Scenario

Context

Hello,
Sometimes in the dataset, target variable is pre-encoded into the 0, 1. But in those sometimes there are other sometimes when those 0 and 1 are encoded in as float. So they are 0.0 and 1.0.

Problem

Now, when we create the PMML for such data the output code is generated like this:

<Output>
	<OutputField name="probability(0.0)" optype="continuous" dataType="double" feature="probability" value="0.0"/>
	<OutputField name="probability(1.0)" optype="continuous" dataType="double" feature="probability" value="1.0"/>
</Output>

So, when we try to predict with other data with the following sentence:

Dataset<Row> result = pmmlTransformer.transform(DF);

The error is generated like:

Exception in thread "main" org.apache.spark.sql.AnalysisException: No such struct field `probability(0.0)` in income_>50K, probability(0.0), probability(1.0);
...

A Fix

So I tried, manually removing the . from the .pmml file and that worked correctly!
The updated code that worked correctly:

<Output>
	<OutputField name="probability(0)" optype="continuous" dataType="double" feature="probability" value="0.0"/>
	<OutputField name="probability(1)" optype="continuous" dataType="double" feature="probability" value="1.0"/>
</Output>

I can understand, that this could be solved from the file where we are generating the PMML but that might not be possible all time. So for the convenience I would ask this community to fix this at the jpmml level.

Thanking you 😄