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/
trapz has joined #mlpack
trapz has quit [Quit: trapz]
kshitijk has joined #mlpack
govg_ has quit [Quit: leaving]
witness___ has joined #mlpack
kshitijk has quit [Ping timeout: 250 seconds]
kshitijk has joined #mlpack
kshitijk has quit [Ping timeout: 246 seconds]
govg has joined #mlpack
kshitijk has joined #mlpack
govg has quit [Quit: leaving]
trapz has joined #mlpack
vedhu63w_ has quit [Ping timeout: 244 seconds]
vedhu63w_ has joined #mlpack
vedhu63w_ has quit [Remote host closed the connection]
witness___ is now known as witness
kshitijk has quit [Ping timeout: 256 seconds]
kshitijk has joined #mlpack
edlinde has joined #mlpack
< edlinde>
where do I get the source code for mlpack?
< edlinde>
is there some sort of a github repo that I can browse?
< edlinde>
there doesn't seem to be a sparse version for LSH
< edlinde>
is anyone aware of a sparse LSH for c++?
< naywhayare>
edlinde: sounds like you found the code (if not, github.com/mlpack/mlpack, or, grab the stable version from mlpack.org)
< naywhayare>
you're right that LSH currently doesn't support sparse matrices
< naywhayare>
it would be pretty easy to refactor, though, I think
< edlinde>
hmm
< naywhayare>
we'd basically need to just templatize the LSHSearch class, adding a 'MatType' parameter; then, replace all 'arma::mat' with 'MatType'
< edlinde>
yeah maybe :)
< edlinde>
you also would need to fix the loading part
< naywhayare>
for the command-line program? yeah
< edlinde>
Is MatType sort of including all sparse matrices CSR, CSC etc?
< naywhayare>
MatType would assume an Armadillo sparse matrix, which is CSC
< naywhayare>
(arma::sp_mat)
< edlinde>
nah I meant that you not only need to fix the LSH search, but also the bit where the LSH is built
< edlinde>
so I would be able to give it a file with say things in libSVM format.. which has sparse entries and then it should be able to read those in
< naywhayare>
that's a question at a higher level than LSH, but yeah, I don't disagree that that would be a good idea
< edlinde>
I haven't looked at the entire code, so I cannot tell
< edlinde>
I was just assuming that you also have a LSHBuild class or something :)
< naywhayare>
nah, for LSH, everything is contained in the LSHSearch class (which can be found in src/mlpack/methods/lsh/lsh_search.hpp and lsh_search_impl.hpp)
< edlinde>
i see .. ok
< naywhayare>
so any dataset loading or any data preparation happens outside of that class
< edlinde>
I saw some ticket which was open for 3-4 years now .. to have sparse versions for all algos
< edlinde>
:)
< naywhayare>
yeah; it would be really nice to get further on that, but time is limited, as always :)
< edlinde>
true that
< edlinde>
:)
< naywhayare>
for some algorithms it's less clear how to templatize to get sparse support, and in some cases we need to wait on Armadillo to support some additional factorizations
< naywhayare>
but in this particular case I think LSH should actually not be too hard