11:57 a.m.
Hello there,
Are there any plans to incorporate any Machine Learning algorithms in Gretl?
Thank you,
PG
*Periklis Gogas
<http://www.econ.duth.gr/personel/dep/gkogkas/index.en.shtml>*
Associate Professor
of Economic Analysis and International Economics
Department of Economics, Democritus University of Thrace
Euro Area Business Cycle Network - Fellow
<http://www.eabcn.org/person/periklis-gogas>
The Rimini Centre for Economic Analysis - Fellow
<http://www.rcfea.org/component/option,com_frontpage/Itemid,1/>
The Society for Economic Measurement - Member
<http://sem.society.cmu.edu/home.html>
Institute for Nonlinear Dynamical Inference (INDI) - Charter Fellow
<http://icemr.ru/institute-for-nonlinear-dynamical-inference/>
12:17 p.m.
Am 20.07.2017 um 13:57 schrieb Periklis Gogas:
Hello there,
Are there any plans to incorporate any Machine Learning algorithms in Gretl?
Thank you,
PG
*Periklis Gogas
Hey Preiklis,
I don't know whether Jack, Allin or anybody else plan to implement such
things into gretl's C-code.
However, it would be nice to have a wrapper, for instance, for the
glmnet-package written by Hastie et al. (URL:
https://web.stanford.edu/~hastie/swData.htm#Glmnet)
For this package there exists already a R, Matlab and Python wrapper. Of
course, the R and Python codes can always be called from gretl's
foreign-environment.
I think the very basic functions of the glmnet package are even written
in Fortran. I am curious whether gretl could call and process Fortran code?
Best,
Artur
12:22 p.m.
Thank you Artur,
That would be very interesting. I am not familiar with that but with the
libsvm.
PG
*Periklis Gogas
<http://www.econ.duth.gr/personel/dep/gkogkas/index.en.shtml>*
Associate Professor
of Economic Analysis and International Economics
Department of Economics, Democritus University of Thrace
Euro Area Business Cycle Network - Fellow
<http://www.eabcn.org/person/periklis-gogas>
The Rimini Centre for Economic Analysis - Fellow
<http://www.rcfea.org/component/option,com_frontpage/Itemid,1/>
The Society for Economic Measurement - Member
<http://sem.society.cmu.edu/home.html>
Institute for Nonlinear Dynamical Inference (INDI) - Charter Fellow
<http://icemr.ru/institute-for-nonlinear-dynamical-inference/>
On Thu, Jul 20, 2017 at 3:17 PM, Artur Tarassow <
artur.tarassow(a)googlemail.com> wrote:
Am 20.07.2017 um 13:57 schrieb Periklis Gogas:
> Hello there,
>
> Are there any plans to incorporate any Machine Learning algorithms in
> Gretl?
>
> Thank you,
>
> PG
>
> *Periklis Gogas
>
Hey Preiklis,
I don't know whether Jack, Allin or anybody else plan to implement such
things into gretl's C-code.
However, it would be nice to have a wrapper, for instance, for the
glmnet-package written by Hastie et al. (URL:
https://web.stanford.edu/~hastie/swData.htm#Glmnet)
For this package there exists already a R, Matlab and Python wrapper. Of
course, the R and Python codes can always be called from gretl's
foreign-environment.
I think the very basic functions of the glmnet package are even written in
Fortran. I am curious whether gretl could call and process Fortran code?
Best,
Artur
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
1:13 p.m.
Am 20.07.2017 um 14:22 schrieb Periklis Gogas:
Thank you Artur,
That would be very interesting. I am not familiar with that but with
the libsvm.
Interesting -- could you explain how familiar you are with libsvm? In
principle this looks like a good candidate to wrap, technically
speaking, because AFAICS it's a C/C++ library.
Apart from that in general I agree with Artur that wrapping some
existing approach in R might also be a good idea. (Assuming that in the
background the respective R package is also calling compiled C/Fortran
code.) I think it is utopian to think that we could or should re-invent
or port everything developed for R.
Obviously for the last couple of years R has had a huge momentum behind
it. I think this is good, even though in some sense we are competition.
cheers,
sven
10 p.m.
On Fri, 21 Jul 2017, Sven Schreiber wrote:
Am 20.07.2017 um 14:22 schrieb Periklis Gogas:
> Thank you Artur,
>
> That would be very interesting. I am not familiar with that but with
> the libsvm.
Interesting -- could you explain how familiar you are with libsvm? In
principle this looks like a good candidate to wrap, technically
speaking, because AFAICS it's a C/C++ library.
I find this interesting too. Periklis, if you've successfully used
libsvm for econometric purposes, please give us a (at least) a brief
description. Libsvm looks like something we could work with quite
easily.
The profile of machine learning has been raised recently by the
article by Mullainathan and Spiess in the Spring 2017 issue of the
Journal of Economic Perspectives, and it's possible this is
something we'd want to explore in gretl.
Allin Cottrell
11:28 a.m.
Hello again, I was away for a few days.
Back to this issue.
I find this interesting too. Periklis, if you've successfully used libsvm
for econometric purposes, please give us a (at least) a brief
description.
Libsvm looks like something we could work with quite easily.
The profile of machine learning has been raised recently by the article by
Mullainathan and Spiess in the Spring 2017 issue of the Journal of Economic
Perspectives, and it's possible this is something we'd want to explore in
gretl.
Allin Cottrell
In my research I have used Support Vector Machines/Regression in
forecasting both financial and macroeconomic variables (I attach two
relevant papers, one on exchange rate forecasting and one on housing prices
). These applications of ML were coded using matlab.
The thing with the SVM/R is to fine-tune the hyperparameters and the kernel
parameters. Libsvm already provides ready wrappers for matlab and R that
are used to import values and export results (available in their page). The
main optimization routine is written in C so it is quite fast. All you need
to build is the wrapper to link libsvm libraries with gretl. Of course you
could also develop you own functions for cross validation training and
parameter seach. That shouldn't be that hard.
It would be nice to be able to use Gretl for Machine Learning!
11:34 p.m.
On Fri, 28 Jul 2017, Periklis Gogas wrote:
[...]
In my research I have used Support Vector Machines/Regression in
forecasting both financial and macroeconomic variables (I attach
two relevant papers, one on exchange rate forecasting and one on
housing prices ). These applications of ML were coded using
matlab.
[...]
It would be nice to be able to use Gretl for Machine Learning!
Thanks, Periklis. We have something you might like that's close to
"proof of concept" stage.
I'll report back once it's a little further along, but it involves a
gretl "plugin" that would link to libsvm and deliver svm
functionality via a gretl function (conditional on libsvm being
available, which we can probably arrange quite easily in our Windows
and Mac packages).
Allin Cottrell
2464
days inactive
2472
days old
6 comments
4 participants
participants (4)
-
Allin Cottrell -
Artur Tarassow -
Periklis Gogas -
Sven Schreiber