Maximum variable size is exceeded when training neural network

Hi,

I've been trying to train a neural network using an input matrix of size 1558x2787, mostly containing boolean values and a target matrix of size 2x2787, the problem is that I'm getting an error saying that "the maximum variable size is exceeded" when I try to train the network! I'm running the student version on 64-bit OS. Can you please help with that? Thanks!

Answer 1

[Greg Heath]

Reduce the number of input variables.

1.You could start by just randomly dividing the N = 1558 dimensional data into M (~10?) disjoint subsets with a smaller No. (~156 ?) of input variables.

2. Next standardize (zero-mean/unit-variance) each subset and use the stepwisefit linear model to choose a smaller subset.

3. Combine the 10 smaller subsets and either repeat the reduction with stepwisefit or proceed directly to designing a net.

Hope this helps.

Greg