Hey, I am using KNIME to do some text mining. After a learning phase with the Naive Bayes Learner, I use a Naive Bayes Predictor in order to evaluate the performance of the predictor. Unfortunately, this predictor fails during execution with the following output:
ERROR Naive Bayes Predictor Execute failed: Most likely class must not be null
All class fields of the documents are correctly filled in. Also, when I exclude certain document rows, the predictor does not fail, but I didn’t manage see a pattern in these excluded documents.
I have fixed it now through removing the Document Vector from my workflow.
I am facing same problem while running naive bayes for text classification. I am unable to understand the fix “I have fixed it now through removing the Document Vector from my workflow.” for it.
My work flow contains two file readers to read training and test data and Naive bayes learner and predictor. predictor is connected to learner and test data. It produces error
"ERROR Naive Bayes Predictor Execute failed: Most likely class must not be null"
the current implementation of the Naive Bayes predictor has indeed a problem with many numerical attributes that have a low variance such as term vectors.
We are looking into the problem and will provide a fix for it.
a fix for this problem will be available in the next release which will be version 2.2.