dataviz for reddit with unsupervised machine learning
	Hi group, as part of our experiments, we applied one of our set of techniques to Reddit, you can see the results and play with it online here: [link] The reddit dataviz is based on a static dataset of ours. enjoy, Em. By Emmanuel Benazera  - Apr 16 - 1 new of 1 message

	A visualization of this data
	Hi Guys, I downloaded the data you have posted here to create a visualization oh how closely different subreddits are linked to one another. Essentially what I did is for every pair of subreddits I looked for if users voted in both more often than would be expected by chance, and if so, made a link (using... more » By Adrian  - Mar 31 - 4 new of 4 messages

	A topic-based subreddit recommender
	Hi all, A few of us at XPLR have released a first version of a subreddit recommender, [link] This recommender builds on top of a clustering algorithm. It is directly usable in your browser and on all Reddit pages, through a user script. See [link] for both code,... more » By Emmanuel Benazera  - Feb 12 - 12 new of 12 messages

	Mahout
	Apache Mahout: Scalable machine learning and data mining [link] This is very relevant to what we're doing since it implements many algorithms which can be used for collaborative filtering: (Full list of algorithms: [link])... more » By Alex Mizrahi  - Apr 7 2012 - 5 new of 5 messages

	plans
	hi So it looks like basic infrastructure for link-level recommendations is ready. It only lacks quality-measuring tool, but it isn't hard to implement. There are things which should be improved (like, a bigger data set), but I think they can wait. I plan to work on infrastructure for subreddit-level recommendations,... more » By Alex Mizrahi  - Apr 4 2012 - 22 new of 22 messages

LSI Test #1 -- Success

So I ran my LSI code on this new data set in a 'proper' fashion: with test-links, test-users and independent check of predictions. So far it looks like it kinda works, although I'm now to sleepy to interpret data properly. Here's data for 100 users. First number is number of votes in test set, second is number of predicted links, i.e. top 100 links, third is... more » By Alex Mizrahi  - Apr 2 2012 - 4 new of 4 messages

	SVD Test #1 - Fail
	Hello Everyone, I've loaded Alex's data into an SVD-based recommender with little success. This is my first SVD-based project and I think I will need to approach it a bit differently. Right now I'm doing the following (using Python with numpy/scipy): Create Matrix A using the training data (links as columns, users as... more » By James Lampton (dorefiend)  - Apr 1 2012 - 9 new of 9 messages

test data set is available

hi I've implemented a simple algorithm which splits whole data set into training and test subsets. All links are split into training and test sets. Users are not split, thus you need to make predictions for all users. Votes for links in training set are represented as is. Votes for links in test set are split into two parts -- "info" part is... more » By Alex Mizrahi  - Apr 1 2012 - 4 new of 4 messages

	some data available
	Hi I've made an extract out of public data, as assembled by Quentin's dumper.py. Only part of data was used (I haven't collected everything yet), but it should be enough for the start. (I'll release update later.) There are 5260381 votes, 2337323 links and 17261 users in this set. [link]... more » By Alex Mizrahi  - Mar 31 2012 - 1 new of 1 message

	issue tracker
	hi Some people asked for issue tracker. Well, it looks like github can do that. (Although I haven't tried it yet so I don't know whether it is convenient enough. If it's not we can try something else. Maybe Trac.) So here it is: [link] It's empty, though. As is repository. Well, somebody has to write... more » By Alex Mizrahi  - Mar 31 2012 - 1 new of 1 message