Matrix dimension problem during training with newff?

In a benchmark dataset if there is a lots of zero in the input file does newff reduces input parameter ?

I have 600 data each with 51 parameter (600 row, 51 column) in a excel file. I devided the dataset 50% for training and rest 50% for testing.

I am havig a problem that is while feeding 51 input during training using newff, after training when i check the network architecture by net.iw it shows 16x49 (16 hidden node & 49 input) instead of showing 16x51 (16 hidden node & 51 input).

the network should be 16x51. otherwise during i am getting matrix dimension error.

anyone know how to solve this ?

Answer 1

[Greg Heath]

>In a benchmark dataset if there is a lots of zero in the input file does newff reduces input parameter ?

Which benchmark dataset?

Inputs, targets and outputs are VARIABLES. Just about everything else is a parameter.

Unlike functions in other toolboxes, the NNTBX variables are in matrix rows and the examples are matrix columns.

The 'remconstantrows' default in NEWFF will remove rows with zero variance. I haven't checked the source code to see exactly how it is done. However it is very possible that variable rows will be removed if the variance is below a threshold. Obtain the source code via

type newff

Otherwise, I don't think any of the neural nets automatically perform variable reduction.

>I have 600 data each with 51 parameter (600 row, 51 column) in a excel file. I devided the dataset 50% for training and rest 50% for testing.

>I am havig a problem that is while feeding 51 input during training using newff, after training when i check the network architecture by net.iw it shows 16x49 (16 hidden node & 49 input) instead of showing 16x51 (16 hidden node & 51 input).

>the network should be 16x51. otherwise during i am getting matrix dimension error.

>anyone know how to solve this ?

[I N ] = size(input) = [ 51 600 ]

If var(input')' has zero rows , the input has constant rows.

Also, rank(input) will be less than 51.

Hope this helps.

Thank you for formally accepting my answer.

Greg