How to train MNSIT digit data using Naive Bayesian classifier (fitcnb)

How to train digitTrain4DArrayData (Inbuilt matlab digit data) using Naive bayesian (fitcnb).

[Xtrain,Ytrain]=digitTrain4DArrayData

Converted 4D data into 2D. ( Tried reading MNIST data from external .csv file also)

inputNaive=fitcnb(inputs,targets)

inputs ( 60000*784) double

targets (6000*1) (tried with double and categorial type)

While training it is showing some error like : A normal distribution cannot be fir for the combination of class 1 and predictor X1. The data has zero variance.

Answer 1

[Puru Kathuria]

Hi Sudheer,

As far as I understand there is no issue with the data or the function fitcnb. The error signifies that a few columns/features might have zero variance. The fit is based on fitting a normal distribution separately for each class and feature.

The within class variance must be positive in the underlying data and If any class has 0 variance for a feature, that normal fit is invaluable.

You might want to use a different model/classifier for this problem.

[Sinan Islam]

Thank you!