09:00 < theJonan> Hi, again. Another question regarding DET - do you have an idea at what sizes of data the DET learning can handle in reasonable time - couple of hours. For example I've tried it with 500k samples of 240 variables and it finished for ≈3 hours (on MacBookPro i7, 8GB Ram). However, the 1.32mln samples with 240 variable or even a 40 variables subset fails to finishes for a whole night.

09:01 < theJonan> All tests are done with 5-fold CV.

14:52 < rcurtin> mentekid: I spent a while with equation 13 of the LSH tuning paper... I wish they had been more verbose there

14:53 < rcurtin> they say "we use the following approximation to simplify computation" but then don't describe how to actually compute that approximation

14:53 < rcurtin> and here I am looking at a whiteboard trying to figure out how to integrate gaussian CDFs...

15:44 < rcurtin> hello everyone who's watching, I wanted to talk about mlpack 2.1.0, which I am hoping to release next week (I know I have said that a lot but I am serious this time! :))

15:45 < rcurtin> I'm planning to merge in PR #800 (singletons) and try to help with the LSH tuning code and see where that ends up

15:46 < rcurtin> I'd also like to look at Tham's csv loading code and give that a good review and see if that will be ready in time

15:46 < rcurtin> but other than that I had nothing specific I was planning to do, so if there is anything I've missed that anyone wants to see in mlpack 2.1.0 let me know and we can make it happen

22:06 < mentekid> rcurtin: Hey

22:06 < mentekid> just saw your message

22:06 < mentekid> I will take a look tomorrow :)

22:17 < rcurtin> yeah, I did not have time to derive any further

22:17 < rcurtin> I saw the lshkit implementation and understood it

22:17 < rcurtin> but I didn't see how their result was the expected value of the definite integral of the gaussian pdf...