PyBrain -- the Python Machine Learning Library | Hacker News Reader

PyBrain -- the Python Machine Learning Library

github.com

39 pointsjnaut11 comments

This is mainly a neural network library, it has svm support mainly for comparison.

It still can't replace weka... which you can kinda use with python http://weka.wikispaces.com/Can+I+use+WEKA+from+Python%3F

This post's scope was about an ML library in Python. There was in fact no such argument made about replacement of any other much more evolved project like Weka, Mallet, etc.

But still, thanks for bringing forth the fact that there are much more sophisticated libraries/pkgs that exist, though in Java, but can be used with Python. Which is actually true for any Java code, i.e. it is accessible in Python, I use JPype for it, among the few other available options for the purpose.

Weka and mallet sound like excellent excuses to learn Scala or Groovy.

This isn't the first time I've seen 'learning,' 'cognitive,' or 'brain' Python libraries come up in on HN. I'm always interested, but can never glean enough from them to understand what to use them for.

Can anyone offer anecdotal experience your machine learning project?

skynet? hello?

I can see in your profile "about:librarian". Hence I am a bit confused as to why you don't know where to use ML, but still, I will go ahead and just scratch the surface for you. Since listing all the fields where ML is/can-be used will require a book by itself. :-)

Well I shall limit my answer to two simple examples from the web and in the vicinity of your profile, so that you find it more interesting.

Did you know that ML is one of the key components of Google(1) search, yes, the one that you might be using day in and day out.

Did you know that all the sites offering you personalized recommendations like "people who bought this book also liked this book", use ML.

Why don't you do a fun experiment. Take up a programming language (I suggest Python) and write an algo (using PyBrain and NN's in it) that reads and learns from habits of the students (book borrowers in this case) and then recommends them on what they should read next, based on what the machine has learned from their past behavior. It won't be perfect and best prediction, rather it will be pathetic at the begining, but it will recommend good and relevant books as time goes by and the machine/algo knows more about the students.

(1)for that matter any good search engine.

Have fun!!

Well!! Did I just give a good product idea?? """A recommendation engine for use in libraries to learn from reading habits of students(borrowers)and hence suggest next reads.""" Not sure how much money can it make though. :-)

Edit: Online book/movie selling/renting engines do that, but not sure of real world libraries doing that. Especially now , when the books are going "e". :-)

Check out goodreads.com.

thanks for sharing the link.

Yeah, there's already plenty of libraries and library-vendors that offer recommendations (Librarything, OCLC's Navigator). Even some open-source online library catalogs have rudimentary recommendations based on subject categories. Libraries likely don't offer recommendations based on a user's checkout history because of a long tradition of utmost privacy concerning a patron's information.

As a librarian though, I'm much more interested in tracking and understanding uses of scholarly journal articles. Most research in academic libraries occurs online, and the current vendors do little but index and provide a search engine.

Good point: """Libraries likely don't offer recommendations based on a user's checkout history because of a long tradition of utmost privacy concerning a patron's information."""

That's where privacy statements like these come in:

http://www.goodreads.com/about/privacy [courtesy j2d2]

Reply on news.ycombinator.com