04:25 < travis-ci> mlpack/mlpack#5460 (master - f829451 : Ryan Curtin): The build was fixed.

04:25 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/2be49aea4d4c...f829451547f9

04:25 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/420378983

15:59 < cult-> sometimes the eigen values are changing signs all together in the matrix while using the pca implementation. why?

16:35 < rcurtin> which solver are you using?

16:35 < rcurtin> which solver are you using?

17:04 < cult-> i am using the defauult exact_svd_method policy

17:05 < cult-> *default

17:07 < cult-> centering doesn't seem to help. which one should i use if i am facing with this kind of error?

17:28 < zoq> The default method is based on SVD which itself provides no means for assessing the sign (the sign is cancel out in the SVD formula), so you are right the signs of the eigenvectors are arbitrary. But I don't see an issue here except that the interpretation isn't that clear.

17:28 < zoq> We could flip the sign based on the largest value of u or v, this is what scikit does to return the same results.

17:32 < zoq> https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/utils/extmath.py#L538

17:32 < zoq> is the function I'm talking about

17:40 < cult-> zoq: what about the other methods?

18:50 < zoq> cult-: randomized SVD does the same, not sure about the quic SVD method; I could take a look into the sign feature if there is need

20:04 < cult-> zoq: that would be awesome!

20:04 < cult-> Otherwise i have to write a work-around.

20:05 < cult-> Please let us know how it proceeds

20:06 < zoq> cult-: Sure, I will look into it tomorrow.

20:07 < cult-> That's fantastic!