 |
|
| Computers Forum Index » Computer Artificial Intelligence - Neural Nets » Paper reference of a heuristic equation for... |
|
Page 1 of 1 |
|
| Author |
Message |
| weiying... |
 Posted: Wed Jul 29, 2009 6:33 pm |
|
|
|
Guest
|
Hi all,
I once read a paper that describes a heuristic equation for finding
optimal number of hidden neurons. it is: the number of hidden units=
(the number of variables + the number of classes )/2. I forget where
it is. If someone know the source of this equation or some related
work, pleas let me know.
Thank you. |
|
|
| Back to top |
|
|
|
| GavinCawley at (no spam) googlemail.com... |
| Posted: Wed Jul 29, 2009 7:27 pm |
|
|
|
Guest
|
On 29 July, 19:33, weiying <luweiyin... at (no spam) gmail.com> wrote:
Quote: Hi all,
I once read a paper that describes a heuristic equation for finding
optimal number of hidden neurons. it is: the number of hidden units=
(the number of variables + the number of classes )/2. I forget where
it is. If someone know the source of this equation or some related
work, pleas let me know.
Thank you.
None of the rules of thumb really work (IMHO). If it is a linear
problem, you don't need any hidden layer neurons, if you have a
pathological problem (try the ringnorm benchmark) you will need a very
large number. I am working on an empirical study on this, I may have
the first set of results within the next month or so. The interim
results suggest using a large network and Bayesian regularisation as a
basic strategy for generalisation, although that is no panacea.
However, first test to see if it is a linearly separable problem.
HTH
Gavin
BTW, I'm sure Greg will be able able to help with this one. |
|
|
| Back to top |
|
|
|
| Greg... |
| Posted: Thu Aug 06, 2009 12:58 am |
|
|
|
Guest
|
On Jul 29, 2:33 pm, weiying <luweiyin... at (no spam) gmail.com> wrote:
Quote: Hi all,
I once read a paper that describes a heuristic equation for finding
optimal number of hidden neurons. it is: the number of hidden units> (the number of variables + the number of classes )/2. I forget where
it is. If someone know the source of this equation or some related
work, pleas let me know.
Thank you.
If someone know the source of this equation, DON'T LET ANYONE
KNOW!
The optimal No. can only be found by trial and error. Rules of thumb
like
Neq >> Nw
are helpful. Search the archives using
greg-heath Neq Nw
Hope his helps.
Greg |
|
|
| Back to top |
|
|
|
| Greg... |
| Posted: Sat Aug 08, 2009 1:11 am |
|
|
|
Guest
|
On Aug 5, 8:58 pm, Greg <he... at (no spam) alumni.brown.edu> wrote:
Quote: On Jul 29, 2:33 pm, weiying <luweiyin... at (no spam) gmail.com> wrote:
Hi all,
I once read a paper that describes a heuristic equation for finding
optimal number of hidden neurons. it is: the number of hidden units> > (the number of variables + the number of classes )/2. I forget where
it is. If someone know the source of this equation or some related
work, pleas let me know.
Thank you.
If someone know the source of this equation, DON'T LET ANYONE
KNOW!
The optimal No. can only be found by trial and error. Rules of thumb
like
Neq >> Nw
are helpful. Search the archives using
greg-heath Neq Nw
Of course if Neq >> Nw is not satified, you can also use
Early Stopping and/or regularization.
See the FAQ.
Greg |
|
|
| Back to top |
|
|
|
| weiying... |
| Posted: Fri Aug 21, 2009 6:52 pm |
|
|
|
Guest
|
On Aug 7, 9:11 pm, Greg <he... at (no spam) alumni.brown.edu> wrote:
Quote: On Aug 5, 8:58 pm, Greg <he... at (no spam) alumni.brown.edu> wrote:
On Jul 29, 2:33 pm, weiying <luweiyin... at (no spam) gmail.com> wrote:
Hi all,
I once read a paper that describes a heuristic equation for finding
optimal number of hidden neurons. it is: the number of hidden units> > > (the number of variables + the number of classes )/2. I forget where
it is. If someone know the source of this equation or some related
work, pleas let me know.
Thank you.
If someone know the source of this equation, DON'T LET ANYONE
KNOW!
The optimal No. can only be found by trial and error. Rules of thumb
like
Neq >> Nw
are helpful. Search the archives using
greg-heath Neq Nw
Of course if Neq >> Nw is not satified, you can also use
Early Stopping and/or regularization.
See the FAQ.
Greg- Hide quoted text -
- Show quoted text -
I have some clues. Thank you all for your suggestions. |
|
|
| Back to top |
|
|
|
| George Campbel... |
| Posted: Sat Aug 22, 2009 1:55 am |
|
|
|
Guest
|
On Aug 22, 2:52 am, weiying <luweiyin... at (no spam) gmail.com> wrote:
Quote: On Aug 7, 9:11 pm, Greg <he... at (no spam) alumni.brown.edu> wrote:
On Aug 5, 8:58 pm, Greg <he... at (no spam) alumni.brown.edu> wrote:
On Jul 29, 2:33 pm, weiying <luweiyin... at (no spam) gmail.com> wrote:
Hi all,
I once read a paper that describes a heuristic equation for finding
optimal number of hidden neurons. it is: the number of hidden units> > > > (the number of variables + the number of classes )/2. I forget where
it is. If someone know the source of this equation or some related
work, pleas let me know.
Thank you.
If someone know the source of this equation, DON'T LET ANYONE
KNOW!
The optimal No. can only be found by trial and error. Rules of thumb
like
Neq >> Nw
are helpful. Search the archives using
greg-heath Neq Nw
Of course if Neq >> Nw is not satified, you can also use
Can you share it with us please.
I am also working on that but have not gotten headway.
Thanks,
Campbel
Quote: Early Stopping and/or regularization.
See the FAQ.
Greg- Hide quoted text -
- Show quoted text -
I have some clues. Thank you all for your suggestions. |
|
|
| Back to top |
|
|
|
| baldrick... |
| Posted: Sat Aug 22, 2009 7:52 am |
|
|
|
Guest
|
Quote: Neq >> Nw
Hope his helps.
Not really!!!
unless I start using matlab I presume, which I am not about to. So I
guess I'll remain in the dark ;-( |
|
|
| Back to top |
|
|
|
| Greg... |
| Posted: Tue Sep 08, 2009 8:24 am |
|
|
|
Guest
|
On Aug 21, 9:55 pm, George Campbel <gcampbel2... at (no spam) gmail.com> wrote:
Quote: On Aug 22, 2:52 am, weiying <luweiyin... at (no spam) gmail.com> wrote:
On Aug 7, 9:11 pm, Greg <he... at (no spam) alumni.brown.edu> wrote:
On Aug 5, 8:58 pm, Greg <he... at (no spam) alumni.brown.edu> wrote:
On Jul 29, 2:33 pm, weiying <luweiyin... at (no spam) gmail.com> wrote:
Hi all,
I once read a paper that describes a heuristic equation for finding
optimal number of hidden neurons. it is: the number of hidden units> > > > > (the number of variables + the number of classes )/2. I forget where
it is. If someone know the source of this equation or some related
work, pleas let me know.
Thank you.
If someone know the source of this equation, DON'T LET ANYONE
KNOW!
The optimal No. can only be found by trial and error. Rules of thumb
like
Neq >> Nw
are helpful. Search the archives using
greg-heath Neq Nw
Of course if Neq >> Nw is not satified, you can also use
Can you share it with us please.
What do you mean by "it"?
Quote: I am also working on that but have not gotten headway.
What do you mean by "that"?
Quote: Early Stopping and/or regularization.
See the FAQ.
Greg- |
|
|
| Back to top |
|
|
|
| Greg... |
| Posted: Tue Sep 08, 2009 8:26 am |
|
|
|
Guest
|
On Aug 22, 3:52 am, baldrick <philbrier... at (no spam) hotmail.com> wrote:
Quote: Neq >> Nw
Hope his helps.
Not really!!!
unless I start using matlab I presume, which I am not about to. So I
guess I'll remain in the dark ;-(
Confused by your response. What, exactly is your problem.
Greg |
|
|
| Back to top |
|
|
|
| baldrick... |
| Posted: Mon Sep 14, 2009 8:20 am |
|
|
|
Guest
|
On Sep 8, 6:26 pm, Greg <he... at (no spam) alumni.brown.edu> wrote:
Quote: On Aug 22, 3:52 am, baldrick <philbrier... at (no spam) hotmail.com> wrote:
Neq >> Nw
Hope his helps.
Not really!!!
unless I start using matlab I presume, which I am not about to. So I
guess I'll remain in the dark ;-(
Confused by your response. What, exactly is your problem.
Greg
Neq >> Nw
What does that mean? Is it matlab notation?
The problem is that I have no idea what this means ;-( |
|
|
| Back to top |
|
|
|
|
|
All times are GMT
The time now is Mon Nov 30, 2009 11:20 am
|
|