Implementation of a multi-class class classfier using logistic regression

Hello,

I want to implement a multi-class classifier using logistic regression, but I don't know the steps for doing that.

I describe the problem below:

• I have five data sets as training set in the form of matrices with the size 600 x 50.
• Each training data matrix belongs to one of the targets 1, 2, 3, 4, 5.
• I add a bias vector as the first column to each of the training data matrices. Now, each matrix has the size 600 x 51

This is the part where I get confused ... . Some people say we need to combine all of the training data matrices in one matrix and use that matrix to train 5 different we vectors. Some people say in addition to that, we need to add another column to the BIG matrix as label. Some others say we need a Softmax function, some people say we don't!

Anyway, I don't know what to do from this part on. I don't know how to calculate the error and decrease it and train a classifier. The only thing I know (hope it is true) is that at the end I will have 5 different W vectors.

Please help me with that.