Model selection returns NaN for naive bayes

Question

I am doing a model evaluation using k-fold cross validation for feature selection where i have 3 large dataset. The problem is that on my first set of data the multinomial n