Input format to the neural network

2012 10 25
1 个回答
回答已采纳
10 次查看（30 天）

0 个投票

Hi everyone,
My question is how's the suitable and available format of neural network training input and target ?
I hope the answer to depend on an example ,for example if my input consists of (10) classes or patterns , and for each pattern there're (5) samples , and a feature vector of length (20) is extracted from each sample . what's the suitable format of the file that contains these training dataset? and what's the format of the input (p) and target (t) to the neural network toolbox in matlab ?
After searching I understand that the feature vector elements for each class is ordered as rows , and the columns express the multiple classes or patterns , am I right ? if no , then correct that to me please , and if yes , I still want to know about the different samples of the same pattern , do I deal with them as separate classes and specify its target separately ? or what ?
please don't answer me like this :
[ I N ] = size(input)
[ O N ] = size(target)
If you don't mind, use my above example numbers instead to understand you perfectly .
Thanks in advance

1 个评论
显示 -1更早的评论 隐藏 -1更早的评论

aurora
aurora 2012-10-25
编辑：aurora 2012-10-25
Is it like this :
- the rows of input matrix = 20 (the feature vector length)
- the columns of input matrix = 10x5 =50 (no. of classes x no. of samples per class)
- the rows of the target matrix = 10 (the number of outputs)
- the columns of the target matrix= 10*5=50 (no. of classes x no. of samples per class)
=> size(input)=[20 50] & size(target)=[10 50]
???
help!

请先登录，再进行评论。

请先登录，再回答此问题。

 采纳的回答

Greg Heath
Greg Heath 2012-10-25

0 个投票

[ I N ] = size(input)= [ 20 50 ]
[ O N ] = size(target)= [ 10 50 ]
If you use PATTERNNET with it's default batch learning algorithm, the order of the individual inputs is irrelevant as long as they are paired with the correct target class vector.
The order that is easiest for me to demonstrate is
target = repmat(eye(10),1,5);
Hope this helps.
Thank you for formally accepting my answer.
Greg

3 个评论
显示 1更早的评论 隐藏 1更早的评论

aurora
aurora 2012-10-26
Thank you Mr.Greg for your response but I still have some inquiries :
1- You replied
" [ I N ] = size(input)= [ 20 50 ]
[ O N ] = size(target)= [ 10 50 ] " does it mean that my numbers are correct for my problem explained in the beginning ?
2- You mentioned PATTERNNET but I don't have it in my R2010a matlab , I'm using newff command how about it ?
3- What do you mean by inputs should be paired with the correct target class ? do you mean that the number of columns (N) is the same in the two input & target matrices? and can I shorten the columns of target matrix because many of them are repeated as each five samples of the same class has the same target vector ??
Thank you
Greg Heath
Greg Heath 2012-10-26
Yes. The numbers in your previous comment are correct.
I think NEWPR is available in 2010a. NEWFIT and NEWPR are just modified versions of NEWFF that are specialized for regression and curvefitting(NEWFIT) or pattern recognition and classification (NEWPR).
If you use the commands
type newpr
type newfit
you can see that a few modifications are made before and after an internal call to NEWFF.
However, you can still use NEWFF if you wish. Nevertheless you should check to see the few modifications that were made in NEWFIT and NEWPR.
Similarly for the new versions:
FITNET replaces NEWFIT
PATTERNNET replaces NEWPR
and
FEEDFORWARDNET replaces NEWFF.
No I meant the obvious: If input column 17 is a class 3 vector then target column 17 should have a "1" in row 3.
No. The number of columns in the input and target matrices must be equal.
aurora
aurora 2012-10-27
Thank you very much all my questions until now is replied by you , I really appreciate your help .
Best Regards

请先登录，再进行评论。

更多回答（0 个）

类别

在 帮助中心 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

产品

标签

Community Treasure Hunt

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

Start Hunting!

Translated by