Data Structure for Neural Network

2 views (last 30 days)
Roberto
Roberto on 11 Nov 2014
Edited: Greg Heath on 13 Nov 2014
Hi all,
my neural network has for example 2 inputs. First input has dimension 3, second input has dimension 2.
For example, if I want to simulate the network with a time series with 3 timesteps with the structure described above, is the following the correct (and standard) form of the data?
X = {[x;x;x] [x;x;x] [x;x;x]; [x;x] [x;x] [x;x]}
And if I want to simulate with 3 concurrent inputs the correct form is
X = {[x x x; x x x; x x x];[x x x; x x x]}
?
Thanks

Answers (1)

Greg Heath
Greg Heath on 13 Nov 2014
1. If you plug numbers into your expressions you will that find one of them is an illegal syntax.
2. I don't understand the difference between what you say you want and an rdinary 5-dimensional input.
Hope this elps.
Thank you for formally accepting my answer
Greg
  2 Comments
Roberto
Roberto on 13 Nov 2014
Edited: Roberto on 13 Nov 2014
Hi Greg, thank you for your answer.
1. I just tried this
X = {[1;1;1] [1;1;1] [1;1;1]; [1;1] [1;1] [1;1]}
Y = {[1 1 1; 1 1 1; 1 1 1];[1 1 1; 1 1 1]}
X =
[3x1 double] [3x1 double] [3x1 double]
[2x1 double] [2x1 double] [2x1 double]
Y =
[3x3 double]
[2x3 double]
It seems ok...
By the way i tried also the function nndata and it returns exactly the data structure i mentioned. In the guide is reported
"Here static (1 timestep) data of 12 samples of 4 elements is created.
x = nndata(4,12)"
And x is a 1x1 cell; the only element is a 4x12 matrix.
2.
In a standard MLP network i'm quite sure there's no difference between standard single 5 dimensional input and 2 vectors (3+2) input but matlab allows multiple input for a reason (example http://i.stack.imgur.com/3G19K.jpg ). For example I guess we can create a parallel feedforward network with multiple MLP working together also each one with different training, in this case multiple inputs is necessary.
By the way, reading the neural network toolbox guide (page 3-46) it seems my input format is correct but i'd still like to have some human feeback about that. So, are
X = {[1;1;1] [1;1;1] [1;1;1]; [1;1] [1;1] [1;1]}
Y = {[1 1 1; 1 1 1; 1 1 1];[1 1 1; 1 1 1]}
the correct and standard form for
x -> a time series of 3 timesteps, each of them composed of 2 input vectors of dimension rispectively 3 and 2
y -> 3 concurrent inputs described as above ??
Please i need a yes/no :)
Greg Heath
Greg Heath on 13 Nov 2014
Edited: Greg Heath on 13 Nov 2014
I didn't notice that you had cells and not matrices.
I am not familiar with using multiple inputs.
Sorry
Greg
PS Pehaps Mark Beale or someone else can help.

Sign in to comment.

Categories

Find more on Deep Learning Toolbox in Help Center and File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!