How can I use the neural network with variable input length ?

Hello, I would like to create a neural network with different input nodes. How can I change the processing function ( Neurons Numbers ) in the hidden layer nodes? To be changed based on the input nodes. Because the length of input not fixed, sometimes be 50 or 122 or 255 or 34 and so on. 80 data input ( 60 for training and o20 for testing ). With one output. 80 data, each having Z parameters (input is Zx80) and Z between 0 to 256. Thank you

Answer 1

[Greg Heath]

Neural networks are functions that yield a specific input/output relation between constant length I-dimensional "I"nput vectors and corresponding constant length O-dimensional "O"utput vectors.

Is your net supposed to implement regression/curve-fitting or classification/pattern-recognition?

What do your input and output vectors represent?

Since input and output vector lengths must be constant, you will have to add zero-valued components to your current inputs to accomplish that.

We can be of more help if you post additional info.

Hope this helps.

Greg

[fathi]

Thanks a lot for your reply //// yes as you said Neural networks are functions that yield a specific input/output relation between constant length I-dimensional "I"input vectors and corresponding constant length O-dimensional "O"output vectors. I have to do binary classification, and I have 60 inputs with variable length ( the max length is 255 - that mean the each input will be from 1 to 255 numbers to NN) // EX. if input one is = 3 4 5 6 3 2 ( just 5 numbers )the input two will be = 8 5 43 2 4 3 4 5 6 3 ( 10 numbers ) not fixed and the accuracy will be decreased if I add zero-valued components to your each current inputs from 60 to accomplish. I try to found any solution to make the ANN learn from a different length of inputs ( is it correct to use incremental learning instead of Batch mode ) if correct, how can I do that, and how the testing will be.

[Greg Heath]

I thought I made it perfectly clear:

BOTH INPUTS AND TARGETS MUST HAVE CONSTANT LENGTH

Hope this helps

Greg