Train validate test

Dear Matlab experts,

I am new to matlab, so i am learning quick. I have a data matrix with mxn with some missing entries. I came across an imputation program which requires training and testing datas. I couldnt understand on how to classify my mxn data into training , testing data. As i was googling on this theory, there is one more vector called validation. So whats the difference between (train, test)1 and (train, validate & test)2. How do i create 1 & 2 using my dataset? please guide me...

[manoraaju]

btw, my matrix mxn is 100x40.

Answer 1

[Walter Roberson]

Some people call the validation set the "test" set, and other people call the "test" set the validation set.

In my project we went through some confusion about this. As best I gathered, the situation became more a "majority rules" than a clear and well-understood definition of which was which.

The training set is the data subset you feed in to your algorithm to decide what parameters and weights to use. You know the actual data class for each of these samples, and are permitted to use that information in developing your algorithm.

One of the other sets is a data subset that you can examine the performance of the algorithm on, and which it is considered acceptable to observe to feed back in to the choices you make. You need some data to test with to determine whether the fit you got for the training set was "over-training" or generalizes well. You know the actual data class for each of these samples, and you are permitted to use that information in determining how well the algorithm did.

The other set is a data subset that you apply the finalized algorithm to, and are not supposed to look at the results to tune your algorithm. You either do not know the actual data class for these samples or you ignore it, and you use your algorithm to predict the data class.