naywhayare 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/
sumedhghaisas has quit [Ping timeout: 252 seconds]
< jenkins-mlpack>
sumedhghaisas: Changing LMF(Latent Matrix Factorization) to AMF(Alternating Matrix Factorization). NMF and CF modules are updted.
cuphrody has quit []
udit_s has quit [Quit: Ex-Chat]
andrewmw94 has joined #mlpack
oldbeardo has joined #mlpack
sumedhghaisas has joined #mlpack
< naywhayare>
oldbeardo: you were looking for me yesterday but I didn't see your message in time
< sumedhghaisas>
naywhayare: Did the commit compile without errors?? I slept after the commit...
< oldbeardo>
naywhayare: yes, look at algorithm 4, step 3(e)
< oldbeardo>
they have mentioned O(log m)
< oldbeardo>
what is the suitable constant for the number of samples?
< naywhayare>
sumedhghaisas: yeah, it built just fine
< naywhayare>
oldbeardo: you can just take log m as the number of samples, that will probably be fine
< sumedhghaisas>
naywhayare: okay thanks...
< naywhayare>
I think their choice of O(log m) is so that the theory works out well
< naywhayare>
so log m is a good default, but probably the user should be able to specify a parameter too
< oldbeardo>
naywhayare: okay, because I was testing out my implementation with a 5 X 8 matrix
< oldbeardo>
log m turns out to be small for that
< naywhayare>
yeah, I imagine that there's a minimum
< naywhayare>
log(5) = 1.6
< naywhayare>
if you picked 2, it might work, but I think QUIC-SVD is aimed at much larger matrices anyway
< oldbeardo>
yes, that makes sense, for the time being I have used 10
< oldbeardo>
10 * log(num_points) that is
< naywhayare>
ok; I guess in this case that means you end up sampling more points than there are
< oldbeardo>
by the way, I have a working Cosine Tree implementation
< naywhayare>
great!
< oldbeardo>
I will send it to you by tomorrow
< naywhayare>
ok, sounds good
sumedhghaisas has quit [Ping timeout: 276 seconds]
Anand has joined #mlpack
oldbeardo has quit [Quit: Page closed]
< andrewmw94>
naywhayare: Am I correct in thinking that the bounds in the binary space tree always assume a euclidean distance metric?
< naywhayare>
andrewmw94: yes, HRectBound assumes a euclidean metric
< andrewmw94>
so will the searches always work correctly? And can I make the same assumption in the R tree?
< naywhayare>
it's not clear how to build a kd-tree (or similar) in arbitrary metric spaces efficiently
< naywhayare>
yeah, it's fine to assume a euclidean metric
< andrewmw94>
yeah, the efficiency had me stumped
< naywhayare>
the cover tree can work with arbitrary metrics, but kd-trees are restricted to euclidean spaces
< naywhayare>
the basic problem is that calculating the centroid in an arbitrary metric space is generally infeasible
< naywhayare>
or, I should be clearer; you can't explicitly calculate the centroid at all
< andrewmw94>
I was wondering how you would do the volume. And I don't think the way it is setup would prune accurately
< naywhayare>
you can calculate the distance between the centroid and some other point in the dataset, but only by calculating the distance between the other point and every point in the node and then taking an average
< naywhayare>
yeah; so it might be possible, but it's probably not worth anyone's time...
< naywhayare>
cooling failure means that mlpack.org is going down for a few
naywhayare has quit [Quit: leaving]
naywhayare has joined #mlpack
< naywhayare>
mlpack.org is down due to a cooling failure... not sure when it will go back up
< naywhayare>
what? webchat is weird, I don't know what I just did
< naywhayare>
anyway, I have no idea if I sent an accidental message to this channel or not. oh well...
< marcus_zoq>
The last message from you: 'mlpack.org is down due to a cooling failure... not sure when it will go back up', so no weird messages ...
< naywhayare>
yeah, so, I sent a PM in webchat, but it didn't open a new window and I mistakenly thought I'd just sent the whole thing to the channel