verne.freenode.net changed the topic of #mlpack to: http://www.mlpack.org/ -- We don't respond instantly... but we will respond. Give it a few minutes. Or hours. -- Channel logs: http://www.mlpack.org/irc/
stephentu_ has joined #mlpack
stephentu_ has quit [Ping timeout: 265 seconds]
jbc_ has quit [Quit: jbc_]
< stephentu> naywhayare: interestingly enough i'm having a hard time getting mosek to converge for MVU
< stephentu> its a tricky sdp
< stephentu> i'm going to try the eta^T eta thingy
< stephentu> to see if it helps
< stephentu> naywhayare: ok i was able to get this
< stephentu> its neat MVU seems to understand the geometry of the swiss roll much better than laplacian eigenmaps :)
< stephentu> now how this is useful
< stephentu> i have no idea
< naywhayare> yeah, this meshes with my vague memory and deduction that MVU is computationally more costly but gives better results
< naywhayare> the ipython notebooks are a nice way to display the experiment
< naywhayare> if you're having trouble getting Mosek to converge then it's not much of a surprise that LRSDP won't either...
< stephentu> ya i love ipython notebooks
vedhu63w has joined #mlpack
< stephentu> naywhayare: you might find this paper interesting
< stephentu> we should implement some of this stuff :)
< stephentu> the proofs are kinda crazy they are saying that SGD with teh low rank factorization is optimization over some matrix manifold
< stephentu> the assertion of theorem 1 is confusing as hell
vedhu63w has quit [Ping timeout: 265 seconds]
stephentu has quit [Quit: Lost terminal]
curiousguy13 has joined #mlpack
vedhu63w has joined #mlpack
aditya95 has joined #mlpack
curiousguy13 has quit [Ping timeout: 240 seconds]
curiousguy13 has joined #mlpack
aditya95 has quit [Ping timeout: 246 seconds]
jbc_ has joined #mlpack
< zoq> I'm curious wouldn't it make sense to disable the bounds checks (ARMA_NO_DEBUG) if we run benchmarks against other libraries?
HeikoS has joined #mlpack
curiousguy13 has quit [Ping timeout: 240 seconds]
curiousguy13 has joined #mlpack
< naywhayare> zoq: when compiling with -DDEBUG=OFF it automatically adds -DARMA_NO_DEBUG: see CMakeLists.txt, line 45
< zoq> naywhayare: ah okay, Thanks!
< naywhayare> the only problem with doing that is that if there's bad code or if a user passes something that will cause it to crash, the program will usually just segfault
< naywhayare> although I guess for the most part, if the user passes a problematic dataset, then if there's a failure it'll usually be in some factorization
< naywhayare> which will still report a failure despite -DARMA_NO_DEBUG
stephentu has joined #mlpack
HeikoS has quit [Quit: Leaving.]
curiousguy13 has quit [Ping timeout: 245 seconds]
curiousguy13 has joined #mlpack
zoq has quit [Ping timeout: 272 seconds]