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 quit [Quit: Lost terminal]
< stephentu>
naywhayare: you might find this amusing
< stephentu>
so this is how you set up a maxcut SDP in mosek, using the *python* api
< naywhayare>
maybe we should turn that into a nice advertisement blog post or something? :)
< stephentu>
naywhayare: not quite there yet, mosek is still much faster
< stephentu>
haha
< stephentu>
turns out writing interior point solvers well is actually hard
< stephentu>
hence being able to start companies
< stephentu>
based on them
Guest45195 has quit [Quit: leaving]
govg has joined #mlpack
govg has quit [Ping timeout: 252 seconds]
govg has joined #mlpack
stephentu has quit [Ping timeout: 252 seconds]
curiousguy13_ has joined #mlpack
dataVinci has joined #mlpack
dataVinci has quit [Ping timeout: 246 seconds]
curiousguy13_ has quit [Ping timeout: 264 seconds]
curiousguy13_ has joined #mlpack
stephentu has joined #mlpack
< naywhayare>
stephentu: I was screwing around with L-BFGS, making LineSearch() void (it fails, just like you said)
< naywhayare>
do you think that your implementation is robust enough to handle most LRSDP problems? (for some definition of "most"...)
< stephentu>
naywhayare: is the question whether or not L-BFGS is robust, or the outer loop augmented lagrangian method is robust?
< stephentu>
i think L-BFGS might be ok, but certainly the augmented lagrangian method is not
< stephentu>
LR-SDP is really hit or miss, and requires hand holding for each new problem
marcus_zoq has joined #mlpack
zoq has quit [Quit: Lost terminal]
marcus_zoq is now known as zoq
< naywhayare>
stephentu: I believe L-BFGS is robust, so I was asking more about LRSDP itself
< naywhayare>
I'd like to keep the hand-holding out of the L-BFGS class itself if possible, so I'm curious if I might get it to converge by just tweaking some L-BFGS parameters
< naywhayare>
if we can make LineSearch() void and remove a bunch of the checks, we can probably accelerate each L-BFGS iteration (maybe not much; not certain)
< naywhayare>
but if keeping it in the state it is in your lbfgs branch is what's necessary to make it reliably converge, that's fine
curiousguy13_ has quit [Ping timeout: 245 seconds]