MultinomialLogisticRegression

[caej...@gmail.com]

Hello,

I've haven't seen any topics regarding MultinomialLogisticRegression. I have used the (binomial)LogisticRegression with good results on input samples of 784 inputs classed as either a 1 or 0. For a test set, it computes the predicted classes and puts them into a 1d, integer vector, eg {1,1, 0,1...}

I tried to do a similar classification on 784 inputs with training classes of 0,1,2,3,4 using MLR and noticed that the computed predicted classes are put into a 2d, double matrix? I was expecting a similar 1d, integer vector array.

Found an example that iterates to generate the model but doesn't provide code for the 'answer' part : http://accord-framework.net/docs/html/T_Accord_Statistics_Models_Regression_MultinomialLogisticRegression.htm

Found an example that produces 2d matrix of answers :
https://github.com/accord-net/framework/blob/master/Samples/Statistics/Classification%20(MLR)/MainForm.cs

I have 2 questions:

1) Should MLR produce a 1d answer vector?
2) If 2d answers are correct, how to use them and can you provide a complete c# example.

Thanks