Re:Milk: Machine Learning Toolkit for Python
by bravura on Tuesday March 15, 2011 @07:12PM
(#2329909)
Could you please comment on why a Python machine learning developer should use your code and not, say, Shogun or scikits.learn? In what use cases would your code be preferable, or dispreferred?
How does your k-means implementation compare to that of scipy's?
Why not push your code as modules into scikits.learn? Their library is designed to be many loosely coupled components.
Re:Milk: Machine Learning Toolkit for Python
by luispedro on Tuesday March 15, 2011 @07:17PM
(#2329929)
Why would you use my code instead of others? None of the packages covers all of machine learning, so it depends on what you're looking for. I focus mainly on supervised learning and kmeans. I want my algorithms to be as scalable as possible too.
Other projects have other priorities/functionality.
I like my interfaces better too, but I might be biased by being so much more familiar with them.
"""How does your k-means implementation compare to that of scipy's?"""
I think that my code is faster: http://bit.ly/e8VOXy and it is probably more scalable.
Scikits.learn has more functionality in certain aspects. So, if you need those, use it. I started milk when there was no scikits.learn. The interfaces are different and they work together here.
That's http://packages.python.org/milk/benchmarks.html if anyone was wondering
(I'm the author of this package).