On Fri, July 21, 2006 20:43, john w wrote:
Yes, I already told Allin and Jack about JMulTi and I also told Markus about
Gretl on JMulti forum.
It's true that thay are both open source but the main problem with
compatibillity is in different programming languages. JMulti is based on
To an extent. In jmulti, the number-crunching routines, which could be the
main area of cooperation, are written in gauss and java is primarily used for
building the user interface. This makes it much more difficult to share code:
java is a loose C++ derivative, while gauss is a matrix interpreted language
whose syntax is very different from C, which is what we use. It may be
possible to share algorithms, though.
In theory, I think it could be possible for jmulti to use our code by using
our shared library. To be clearer: gretl can be thought of as 3 building
blocks (I'm simplifying things quite a bit here; Allin, correct me if I'm
wrong). The library, which contains all the numerical methods, and the two
clients (graphical and command-line), which allow the user to run the code in
the library through a user interface. For example, when you run OLS via the
Model menu, the gretl program you're running sends all the necessary info to
the library, which returns an object containing the estimated coefficients,
auxiliary statistics and so on. The program's job is to display the results to
you. In principle, anyone could write a program (in C, but also C++, java,
python or whatever) with a different user interface which uses gretl's
library. It should be technically possible to use the library from within a
java program (I don't know how hard it would be, though), so jmulti could
use our methods any time. IIRC, jmulti is GPL'ed too, so there shouldn't be
any legal issues[*]. The reverse would be much harder, because there's no way
I know of of running gauss code from within a C application.
I use them both. The oconnection between them is in Excel files which
Gretl is better in OLD, Logit, Probit, SURE etc. analisys while JMulTi is
better in time series analisys like VAR and VEC, subset modelling.
I would say that jmulti is specialised to time-series analysis, while gretl
aims to be more comprehensive. For example, we're currently fine-tuning a
totally revamped framework for panel data analysis. I don't think the jmulti
people would be even remotely interested in such a thing.
[*] Not really my business, but I wonder what the legal status of jmulti is,
as they're using a non-free app (gauss) as an essential component of a GPL
project. But maybe I misunderstood the whole thing. And, as I said, it's none
of my business.
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"