Surprise – a simple recommender system library for Python | Hacker News Reader

Surprise – a simple recommender system library for Python

surpriselib.com

89 pointsdanso16 comments

GPLv3 will kill this project.

this is a library, and should have a library license, at least LGPLv3.

Looks like the author is a student, so it could be that wide adoption isn't the objective. It's still an excellent resume builder.

Hi, I'm Nicolas, Surprise's author. Thanks for the feedback!

You're not the first person to tell me GPL may not be a good idea and to be fair I never really thought it would make such a difference. Appart from LGPL, what other license would you recommend?

most python based web applications are using Django, and Django is BSD licensed.

https://github.com/django/django/blob/master/LICENSE

a recommendation engine is great for all kind of web apps, and licensing it with BSD makes it more adopted by django web apps.

Thanks,

Thanks for the input, I changed the license to BSD!

Be aware that if you license your work under the BSD then someone else can take your work, change it, improve it, sell it — and never share his changes with you.

So Netflix could take your recommender system, use it to make millions and do nothing but mention you in a footnote somewhere on their website.

And they wouldn't even need to do that.

What is the general approach to doing this in realish time?

The benchmarks on this make than untenable, but you could probably RYO KNN and be in the low seconds ... that still won't work.

Is the approach to build generalized profiles ahead of time and then compare against those in real time?

Are there any actually free/libre recommendation data sets? All that I can find, including the ones used by Surprise (MovieLens, etc.), look to be under restrictive licenses.

My first instinct was to point you to a colleague's page, Justin Grimmer, who does analysis of free-form text such as Congressional press releases (and other statements). I swear he had direct links to zip files of this bulk text data but can't find them on a quick glance. Here's his research listing just to give an idea of what he's able to analyze from this public data: http://www.justingrimmer.org/research.html

As for why there aren't newer, or at least more of a variety of public datasets as canonical as the iris, diamond, Titanic survivors, MovieLens, etc...I'm baffled too (being a recent newcomer to academia).

If scraping is difficult for you, Socrata has made finding data exponentially easier than what it was a few years ago. Here's a great place to start: http://www.opendatanetwork.com/

I can program so scraping is pretty easy. Finding data sets that aren't restricted is the hard part.

Thanks for the links.

If you have an interest in U.S. politics, check out the github/unitedstates repo, which is full of data and assets, particularly about legislators and their votes:

https://github.com/unitedstates

Sunlight Labs, the tech arm of the Sunlight Foundation, is sadly closing down, but they passed many of their projects to ProPublica. Here's a writeup of that to give an idea of what the projects contained (in terms of data): https://www.propublica.org/nerds/item/sunlight-labs-takeover...

Just in case you don't care about the U.S., it must be mentioned that the UK has the best country-level data repository: https://data.gov.uk/

Perhaps HN can use this to recommend stories (?)

(for example, under a "recommended" menu item)

Or something for WordPress :)

Or something for webbrowsers in general :)

But then you need to share all your urls, not sure if everybody would be comfortable with that.

Reply on news.ycombinator.com