Thread Subject: Neural Networks

Hello,
My problem with NN is that simulation values are very high for non trained objects as that for correct simulation.

On Mar 23, 3:00=A0pm, "Enas Moh" <engenas_mas...@yahoo.com> wrote:
> Hello,
> My problem with NN is that simulation values are very high for non traine=
d objects as that for correct simulation.

Need details:

Classification or regression
newff or newrb
Number of observations
size(p)
size(t)
train/validation/testing split
Input Dimensionality
Output dimensionality
Number of hidden nodes
hidden activation function
output activation function
training function
MSE0 =3D mse(t'-mean(t'))
MSEtrn/MSE0
MSEval/MSE0
MSEtst/MSE0

Greg

> > Hello,
My problem with NN is that simulation values are very high for non trained
objects or with wrong decision as that for correct simulation.

object recognition
newff is used
using (10) captures for each of (3) different objects, each has a simulation value between 0,1

using one hidden layer having (10) neurons
output layer has (3) neurons
the (2) layers use (logsig)

traingdx and sse are used

also how to go say from line (2) in m-file after certain condition to line (5) for ex

On Mar 23, 5:14 pm, Greg Heath <he...@alumni.brown.edu> wrote:
> On Mar 23, 3:00 pm, "Enas Moh" <engenas_mas...@yahoo.com> wrote:
>
> > Hello,
> > My problem with NN is that simulation values
> > are very high for non trained objects as that for correct simulation.
> > using (10) captures for each of (3) different objects,
> > each has a simulation value between 0,1

I don't know what that means.

> > also how to go say from line (2) in m-file after certain
> > condition to line (5) for ex

line(2) if (~conditon)
line(3) whatever
line(4) end
lin3(5) blah blah

If you want more help, please fill in the blanks(?)

> Classification or regression

object recognition

> newff or newrb

newff is used

> Number of observations

?

> size(p)

[ 10 ?]

> size(t)

[3 ?]

> train/validation/testing split

Ntrn = ?
Nval = ?
Ntst = ?

> Input Dimensionality

10

> Output dimensionality

3

> Number of hidden nodes

10

> hidden activation function

logsig

> output activation function

logsig

> training function

traingdx

> objective function

sse

> SSE0 = sse(t'-mean(t'))

?

> SSEtrn/SSE0

?

> SSEval/SSE0

?

> SSEtst/SSE0

?



Greg

Dear Sir,

1) what is the difference between regression and classification in NN?
2) what I must write if i want to jump from line (3) to line (6) "after certain condition" in these commands:
line (1): if…..
line (2): whatever
line (3): end
line (4): whatever
Line(5): whatever
Line(6): whatever
3) what is meant by those sentences:
      handles.S = S;
      guidata(hObject, handles);

                                                Thanks in Advance

On Mar 25, 1:27=A0pm, "Enas Moh" <engenas_mas...@yahoo.com> wrote:
> Dear Sir,
>
> 1) =A0 =A0 =A0what is the difference between regression and classificatio=
n in NN?

Always look for definitions in Wikipedia and Wolfram before
asking for them in a newsgroup.

You mean in statistics. Classical Regression involves designing a
mathematical model to aproximate a function y =3D f(x) given only
a sample with paired I/O observation (x1,f(x1)+n1,(x2,y(x2)+n2),
...etc where the ni s are identically and independently distributed
 random variables with zero mean and finite variance.

Classification is a special case of regression where f(x) is a
discrete
index indication the input came from one of c classes. Your object
recognition project falls into this category.

> 2) =A0 =A0 =A0what I must write if i want to jump from line (3) to line (=
6) "after certain condition" in these commands:
> line (1): if=85..
> line (2): whatever
> line (3): end
> line (4): whatever
> Line(5): whatever
> Line(6): whatever

See the online documentation

doc if
help if

> 3) =A0 =A0 =A0what is meant by those sentences:
> =A0 =A0 =A0 handles.S =3D S;
> =A0 =A0 =A0 guidata(hObject, handles);

See the online documentation

doc guihandles
help guihandles
doc guidata
help guidata

Hope this helps

Greg

Thanks in advance for your reply
I read the documents for "if" command, but my question was related on how to go to a certain command in the matlab file changing by this way the normal sequence in the m-file.
I think as " go to line ..." in C language.
                                          Thanks

On Mar 27, 5:14=A0am, "Enas Moh" <engenas_mas...@yahoo.com> wrote:
> Thanks in advance for your reply
> I read the documents for "if" command, but my question was related on how=
 to go to a certain command in the matlab file changing by this way the nor=
mal sequence in the m-file.
> I think as " go to line ..." in C language.
> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 Thanks

Change your way of thinking. GO TO results in
sphaghetti code. That is why it is banned from
mattlab.

help while
doc while

help switch
doc switch

Hope this helps.

Greg

Dear sir
After NN training using error goal "net.trainParam.goal = 1e-6", i get a message that "min gradient is reached, performance goal is not met", is this a problem in NN?
                                         Thanks