mgashler ( on January 14, 2009, 19:43:30 ) Recently I've become interested in the idea of putting together a single archive of open-source cross-platform command-line tools for data mining tasks. (This would obviously be a subset of the tools found on this site.) I could simply go through the collection on this site to gather them, but I was wondering if others might be interested in this effort? Specifically, if we were to announce a call for such tools, I suspect that some developers might be interested in adding command-line interfaces for their tools so that they can be included. Why I'm interested in command-line interfaces: 1- It's common. 2- It's language agnostic. (Tools in C++, Python, Java, Octave, etc. can all be equipped with a command-line interface.) 3- It's fairly script-friendly. I think it is clear that there is no other interface on which everyone will agree. I know "your favorite high-level language" can produce cleaner/shorter/more-intutive scripts than Bash, but you can call command-line tools from your favorite language too. Command-line tools cater to every scripting language. When a new scripting language becomes popular, it will be able to call command-line tools too. Why do we need a common interface for data mining tools? Because there are so many tools.

mgashler ( on February 9, 2009, 23:05:39 ) *crickets chirping* Hmm, maybe I'm the only one who thinks command-line interfaces are the way to make ML tools interact. I'd be interested to hear some rebuttals. What do you think is the best interface to unify ML tools? Python? Matlab? Just be happy doing it all manually insdie a GUI? Or is trying to make various tools all work together a pointless effort?

sonne ( on February 9, 2009, 23:15:15 ) +1 for python but I guess I am extremely biased. If you want I could provide a script to learn/train any of the current cmdline svm-solvers (like libsvm/svmperf/svmlin svmlight etc) as a start :)

Post a Reply