Re: can NN have variable number of features

On Jun 4, 10:29 pm, saeed <weareislamsold...@xxxxxxxxxxx> wrote:
Hi Greg>Explain your problem and I will tell you if that makes >sense to
me.

How many inputs (120?) and outputs (1?)

i'm using it for machine translation
were 120 is the number of word and not all sentences have 120 words
so if the sent had 5 words i do fill(pad) the empty space with 0.1
value

You didn't answer the question.

No. of network inputs =
No. of network outputs =

What do the inputs and outputs represent?
words in sent

Why do your outputs mean the same thing as your inputs?

Why would any inputs be missing?
no

Inappropriate answer.

How much data do you have for training/validation/testing

i just have training/testing
300 for training 10 for testing

You may not have enough data. How many hidden nodes do
you have? i.e., if you are considering a 120-H-1 MLP, how large is
H?

Do you mean convergence?

yes

Convergence has nothing to do with testing.
I don't understand

how come that Convergence has nothing to do with testing.

Testing does not involve iteration.

then how do u make sure that ur NN has learnt something and why they
call testing then??

Testing involves obtaining an output given a nonlearning input.
Iteration and convergence are not involved.

1. what you trying to do
2. what you are doing

explained above.

Not very well.

Please try again.

Hope this helps.

Greg
********************************************



Greg Heathwrote:

On Jun 4, 4:14 am, saeed <weareislamsold...@xxxxxxxxxxx>
wrote:

You didn't answer my question:

"It would be helpful if you explained your problem and why you need
to
do this."

My input has max no 120
and sometimes the data are 7 or 10 in some rows
so what i did is to pad the rest of 120 by .1

Explain your problem and I will tell you if that makes sense to me.

How many inputs (120?) and outputs (1?)

What do the inputs and outputs represent?

Why would any inputs be missing?

How much data do you have for training/validation/testing

that's why i was asking about varaible features as input
in NN
i don't have missing data but they are short than the max

Why?

want to ask is my approach of padding is right
coz i get converion

Do you mean convergence?

in the training but in testing i don't get it at
all

Convergence has nothing to do with testing.
I don't understand

1. what you trying to do
2. what you are doing

Hope this helps.
It would be helpful if you explained your problem and why you
need
to do this.

Your approach depends on your ultimate goal.

Typically, the number of inputs is fixed and missing inputs are
represented by a default value that depends on the type of
problem
you have. Two examples are minimum or average values.

In some cases, missing values are estimated by imputation. If
the expected number of missing values is more than one,
imputation gets complicated.

Again, it depends on what you are trying to do.

Hope this helps.

Greg Heathwrote:

On May 29, 7:09 am, saeed

<weareislamsold...@xxxxxxxxxxx>



wrote:
Hi all

i want to ask if NN can have variable number of
features.
How does it handle it, what value are inserted and who
does it
represent this variables.
Any example of such work

It would be helpful if you explained your problem and why
you
need
to do this.

Your approach depends on your ultimate goal.

Typically, the number of inputs is fixed and missing
inputs are
represented by a default value that depends on the type of
problem
you have. Two examples are minimum or average values.

In some cases, missing values are estimated by imputation.
If
the expected number of missing values is more than one,
imputation gets complicated.

Again, it depends on what you are trying to do.

Hope this helps.

Greg- Hide quoted text -

- Show quoted text -- Hide quoted text -

- Show quoted text -- Hide quoted text -

- Show quoted text -


.

Relevant Pages

  • Re: LCA: model failed to determine a valid solution...why?
    ... With a 1-class model (and no missing data) estimation really needs just ... one iteration, so convergence shouldn't be an issue there. ... Have you experimented with changing the convergence criterion ...
    (sci.stat.edu)
  • Re: LCA: model failed to determine a valid solution...why?
    ... With a 1-class model (and no missing data) estimation only needs one ... iteration, so convergence shouldn't be an issue there. ... Have you experimented with changing the convergence criterion and/or ...
    (sci.stat.edu)
  • Re: Why is the tetration so much harder to extend than the factorial?
    ... But the reason for its failure in fractionally iterating with ... prove convergence? ... framework can be derived ... functional iteration and fractional tetration and can include ...
    (sci.math)
  • Re: Old chestnut: f(f(x))=exp(x)
    ... convergence*, and possibly some other things not rigorously proved!). ... we have a way of expressing the general iteration of exp- 1 as ... a power series, subject to the convergence provisos. ...
    (sci.math)
  • Re: Richardson-Lucy deconvolution ?
    ... corresponding pixel in the observed image, and p_is the ... i.e. the fraction of the pixel at j observed at i.. ... There are two widely used ways to initialize the RL iteration. ... Regarding the convergence, you could check some norm of the difference ...
    (sci.image.processing)