GMDH NETWORK CONFUSED??
akshay bhat
could not understand a thing (given i know considerable about mlp and
rbf) i couldn't get any pseudo algorithm. nor any toolbox in matlab.
however the main question which i want to ask is that whether gmdh is
a good method or not because i have hardly seen any publications about
it and all listed on its site are 2002 or older.
can anyone tell me whether they are worth knowing or should i learn
support vector machines instead.
Greg Heath
Learn SVM first.
Hope this helps.
Greg
fr...@knowledgeminer.net
look at regression, modeling, or time series problems. also, if you
want to have an analytical expression of your model on the fly, i.e.,
a regression equation, or if you do not know exactly which are your
most relevant input variables to use, or how the topology/structure of
you model should look like, gmdh is worth to look at. it does the work
for you. for more info have a look at
frank
A.L.
wrote:
Depends where you are and what you do. This is very good method.
Unfortunately, not popular in the USA where science is not governed by
scientific mertit, but by political correctness, ability to get
funding (a.k.a. grants), fashion and greed.
A.L.
Phil Sherrod
On 20-May-2007, A.L. <fe...@2005.com> wrote:
> Depends where you are and what you do. This is very good method.
> Unfortunately, not popular in the USA
Well, what country is GMDH popular in? I wasn't aware that there are
geographic boundaries for math models.
> not popular in the USA where science is not governed by
> scientific mertit, but by political correctness, ability to get
> funding (a.k.a. grants), fashion and greed.
I don't know what's behind that gratuitous comment, but I'll be happy to
compare the number of Nobel prizes in math, physics, chemistry and medicine won
by USA scientists with those of any other country. There sure do seem to be a
lot of people from all over the world trying hard to get into USA technical
universities. I guess they're just interested in political correctness and
fashion.
--
Phil Sherrod
(PhilSherrod 'at' comcast.net)
http://www.dtreg.com (Decision trees, Neural networks and SVM modeling)
http://www.nlreg.com (Nonlinear Regression)
A.L.
<PhilS...@REMOVETHIScomcast.net> wrote:
>
>On 20-May-2007, A.L. <fe...@2005.com> wrote:
>
>> Depends where you are and what you do. This is very good method.
>> Unfortunately, not popular in the USA
>
>Well, what country is GMDH popular in? I wasn't aware that there are
>geographic boundaries for math models.
>
Yes, there are boundaries. For example, dominating language in
Artificial Intelligence in Europe was (an is active) Prolog, in USA
was (and is dead) Lisp. For quite a long while dominating computer
language in Europe was Algol 60, in USA Fortran. There is generally
different approach towards attacking problems: analysis in Europe,
brute force in USA. Take Edison (USA) and Tesla (Europe). This is the
best example to illustrate the difference.
>> not popular in the USA where science is not governed by
>> scientific mertit, but by political correctness, ability to get
>> funding (a.k.a. grants), fashion and greed.
>
>I don't know what's behind that gratuitous comment, but I'll be happy to
>compare the number of Nobel prizes in math, physics, chemistry and medicine won
>by USA scientists with those of any other country. There sure do seem to be a
>lot of people from all over the world trying hard to get into USA technical
>universities. I guess they're just interested in political correctness and
>fashion.
Being professor at two U.S. universities for about 12 years and
several universities in Europe for about 20 years. I believe that I
have better comparative material than you have.
Regarding so called "American Noblists" - please check their names and
places of birth.
Regarding quote: "seem to be a lot of people from all over the world
trying hard to get into USA technical universities" - good for US
universities. According to AAAS, only 1% of high school graduates in
US will study science and engineering, compared to 55% in China.
Without foreign students most graduate schools would be forced to
close the door.
However, 10 years ago, majority of these students were doing
everything to stay in the US and to get green card. Today, majority of
them are not interested. They are going back to home countries where
they can get better salaries, better status and no problems. And can
get their Nobel prices. And According to AAAS, USA has already lost
the leadership in science and technology. Including the number of
Noblists.
A.L.
Phil Sherrod
On 20-May-2007, A.L. <fe...@2005.com> wrote:
> >> Depends where you are and what you do. This is very good method.
> >> Unfortunately, not popular in the USA
> >
> >Well, what country is GMDH popular in?
You didn't answer my question: What country is it popular in?
> Yes, there are boundaries. For example, dominating language in
> Artificial Intelligence in Europe was (an is active) Prolog, in USA
> was (and is dead) Lisp. For quite a long while dominating computer
> language in Europe was Algol 60, in USA Fortran.
I don't see your point. GMDH could be programmed in any number of languages.
Are you suggesting because C++ is the dominant programming language in the USA
that that somehow precludes GMDH?
> There is generally
> different approach towards attacking problems: analysis in Europe,
> brute force in USA. Take Edison (USA) and Tesla (Europe). This is the
> best example to illustrate the difference.
You obviously have a real chip on your shoulder about the USA. What is your
nationality?
> Being professor at two U.S. universities for about 12 years and
> several universities in Europe for about 20 years. I believe that I
> have better comparative material than you have.
>
> Regarding so called "American Noblists" - please check their names and
> places of birth.
American is a country of immigrants. Why do you think they came to the USA to
do their work?
Gregory
wrote:
>
> You didn't answer my question: What country is it popular in?
Welcome the GMDH Conference in Prague in September and you'll see how
well the GMDH approach is distributed around the world. As I know it's
still possible to send abstract of the report.
Currently the theory is developing in Ukraine (the origin of GMDH),
China, Iran, Czech Republic, Japan, France, Russia, Canada and Fiji
islands. Here in UK there are several universities which investigate
GMDH type networks. Maybe someone know more.
The new software systems were created in Germany, Czech Republic, UK
and Iran. Possibly it works in stock market forecasting systems in
USA.
GMDH is very interesting and unique approach because it gives
possibility to find the optimal structure of the model or network
during sorting, guarantee that this model will be the most accurate
and solve automatically many complex problems such as overfitting,
forecasting for small number of observations, fixing of needed
variables in the output, take into account the level of noise in your
sample and so on.
There are books at http://gmdh.net/articles which describe this.
Hope to meet you at IWIM-2007.
Regards,
Gregory Ivahnenko
A.L.
wrote:
>On May 21, 6:02 am, "Phil Sherrod" <PhilSher...@REMOVETHIScomcast.net>
>wrote:
>>
>> You didn't answer my question: What country is it popular in?
>
>Welcome the GMDH Conference in Prague in September and you'll see how
>well the GMDH approach is distributed around the world. As I know it's
>still possible to send abstract of the report.
>
>http://www.iwim2007.org
>
>Currently the theory is developing in Ukraine (the origin of GMDH),
>China, Iran, Czech Republic, Japan, France, Russia, Canada and Fiji
>islands. Here in UK there are several universities which investigate
>GMDH type networks. Maybe someone know more.
>
Add Poland. I remember several interesting contributions from Japan,
especially in the area of real time, adaptive GMDH algorithms.
Application domain for these works were environmental problems,
especially water quality modelin gane forecasting
A.L.
A.L.
<PhilS...@REMOVETHIScomcast.net> wrote:
>
>On 20-May-2007, A.L. <fe...@2005.com> wrote:
>
>> >> Depends where you are and what you do. This is very good method.
>> >> Unfortunately, not popular in the USA
>> >
>> >Well, what country is GMDH popular in?
>
>You didn't answer my question: What country is it popular in?
>
Do the research yourself. Go to GMDH web page.
>> Yes, there are boundaries. For example, dominating language in
>> Artificial Intelligence in Europe was (an is active) Prolog, in USA
>> was (and is dead) Lisp. For quite a long while dominating computer
>> language in Europe was Algol 60, in USA Fortran.
>
>I don't see your point. GMDH could be programmed in any number of languages.
>Are you suggesting because C++ is the dominant programming language in the USA
>that that somehow precludes GMDH?
This is not about LANGUAGES. This is about SCIENCE. Sorry that you
don't catch.
>
>> There is generally
>> different approach towards attacking problems: analysis in Europe,
>> brute force in USA. Take Edison (USA) and Tesla (Europe). This is the
>> best example to illustrate the difference.
>
>You obviously have a real chip on your shoulder about the USA. What is your
>nationality?
>
No, I don't have anything negative about USA. I have something
negative about EDUCATION, on all possible levels. The quality is
unbelievably low and substandard. My nationality has nothing common
with my opinion.
>> Being professor at two U.S. universities for about 12 years and
>> several universities in Europe for about 20 years. I believe that I
>> have better comparative material than you have.
>>
>> Regarding so called "American Noblists" - please check their names and
>> places of birth.
>
>American is a country of immigrants. Why do you think they came to the USA to
>do their work?
Oh, yes.... America can BUY whatever is available. This was working
well in the past. But currently, less and less is for sale....
A.L.