NaN results for F1 Score, Precision and Recall

[Vinicius Figueiredo]

I am using MOA GUI with the Morris Dataset 3: Gas Pipeline (http://www.ece.uah.edu/~thm0009/icsdatasets/gas_final.arff) and I am getting the following results:

Does anyone know the reason for the NaNs results in the F1Score, Precision, and Recall?

I executed with three classifiers trying to understand the results. I want to compare these results with the WEKA, but I can not find an explanation of NaN results.

[Vinicius Figueiredo]

I have an update about the subject.

I created a small version of the dataset with 262 samples (50% Normal and 50% Atack) to run with the following script:

EvaluatePrequential -l bayes.NaiveBayes -s (ArffFileStream -f (C:\...\gas_final_part50.csv)) -e (WindowClassificationPerformanceEvaluator -o) -i 262 -f 1 -q 1

At the end I could calculate the following metrics manually:

                         Acuracy     F1Score    Precision    Recall

Naive Bayes    96.56%        96.50%       98.41%    94.66%

The MOA GUI as showing the following metrics:

                         Acuracy     F1Score    Precision    Recall

Naive Bayes    96.56%          ?                   ?                ?

I conclude that the GUI MOA is not giving correct results for F1Score, Precision, and Recall metrics.

Is there a reason for that "?" or "NaN" results? Is it a program bug or am I doing something wrong?

If someone wants to make the same test I can give the reduced dataset I made.