Where is the border between SciPy and ML algorithms?
![](https://secure.gravatar.com/avatar/5d16f696b7bb87053189ad7992db3667.jpg?s=120&d=mm&r=g)
Hi, the scipy.cluster module up to now contains already algorithms for Vector Quantization and Kmeans. In the docs it's said that SOMs are following. Now, after having followed the mails mentioning Orange and Elefant: Where do we draw the line? I'm asking, because a student of mine has implemented SWIG-Python-wrappers to the C++ implementations of diverse decision trees (continuous VFDT, C4.5 and some others), which work quite well. We would of course be glad if those would be included in standard SciPy. However, I'm not quite sure, what the current policy of the core SciPy developers is regarding non-core algorithms. Thanks in advance for any light on this topic, wr
![](https://secure.gravatar.com/avatar/60e03bd1fd9f2dbc750e0899b9e7e71d.jpg?s=120&d=mm&r=g)
Hi, As there are SWIG files and as Scipy does not depend ATM on SWIG, you could create a scikit with or see with David Cournapeau how it could fit in its machine learning scikit. Matthieu 2007/11/8, Willi Richert <w.richert@gmx.net>:
Hi,
the scipy.cluster module up to now contains already algorithms for Vector Quantization and Kmeans. In the docs it's said that SOMs are following. Now, after having followed the mails mentioning Orange and Elefant: Where do we draw the line?
I'm asking, because a student of mine has implemented SWIG-Python-wrappers to the C++ implementations of diverse decision trees (continuous VFDT, C4.5and some others), which work quite well. We would of course be glad if those would be included in standard SciPy. However, I'm not quite sure, what the current policy of the core SciPy developers is regarding non-core algorithms.
Thanks in advance for any light on this topic, wr _______________________________________________ SciPy-user mailing list SciPy-user@scipy.org http://projects.scipy.org/mailman/listinfo/scipy-user
-- French PhD student Website : http://miles.developpez.com/ Blogs : http://matt.eifelle.com and http://blog.developpez.com/?blog=92 LinkedIn : http://www.linkedin.com/in/matthieubrucher
![](https://secure.gravatar.com/avatar/9820b5956634e5bbad7f4ed91a232822.jpg?s=120&d=mm&r=g)
Willi Richert wrote:
Hi,
the scipy.cluster module up to now contains already algorithms for Vector Quantization and Kmeans. In the docs it's said that SOMs are following. Now, after having followed the mails mentioning Orange and Elefant: Where do we draw the line?
My understanding is that we do not throw everything we can think of in scipy; scipy.cluster was kept in scipy for backward compatibility ? That being said, I don't remember having seen any mention of line between core and non core algorithms. Maybe the main difference between scikits and scipy is the license: whereas scipy avoids any non BSD -like license, scikits is freeer in this respect (you can depend on GPL code, for example).
I'm asking, because a student of mine has implemented SWIG-Python-wrappers to the C++ implementations of diverse decision trees (continuous VFDT, C4.5 and some others), which work quite well. We would of course be glad if those would be included in standard SciPy. However, I'm not quite sure, what the current policy of the core SciPy developers is regarding non-core algorithms.
If this is under open source license, as Matthieu said, we (I) can import it into the learn scikits, which implements already several machine learning algorithms (EM for Gaussian mixtures, datasets, SVM). cheers, David
participants (3)
-
David Cournapeau
-
Matthieu Brucher
-
Willi Richert