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/
Anand has joined #mlpack
Anand has quit [Ping timeout: 246 seconds]
marcus_zoq has quit [Ping timeout: 260 seconds]
marcus_zoq has joined #mlpack
< jenkins-mlpack>
Starting build #2015 for job mlpack - svn checkin test (previous build: SUCCESS)
< naywhayare>
the basic idea is to have PointSelectionPolicy able to either select points or provide new points, and have this knowledge available at compile-time
< naywhayare>
then the compiler can provide a NystroemMethod implementation that is as fast as the existing implementation if the point selection policy selects points, or an implementation that's almost as fast if the point selection policy gives new points
< jenkins-mlpack>
Starting build #2017 for job mlpack - svn checkin test (previous build: SUCCESS)
< naywhayare>
I don't think we'll be able to use KMeans<> as a PointSelectionPolicy directly, but it would be easy to provide a wrapper class for KMeans (or any clustering algorithm that provides cluster centroids)
< naywhayare>
sumedhghaisas: were you still having an issue with figuring out how to tell the user whether or not termination was successful or hit the maximum number of iterations?
< naywhayare>
or did you come up with a neat idea?
< sumedhghaisas>
naywhayare: sorry I was little busy this weekend... yes actually I have...
< sumedhghaisas>
one solution I tried is returning a boolean instead of double in amf::Aplly()
< sumedhghaisas>
but I am not really satisfied with that solution...
< naywhayare>
hm, so here was my thought:
< naywhayare>
there is already AMF::TerminationPolicy() which returns the instantiated termination policy
< sumedhghaisas>
yes... right..
< naywhayare>
maybe a user can access that; (AMF.TerminationPolicy().Iterations() == AMF.TerminationPolicy().MaxIterations()) could tell the user if the maximum number of iterations was reached
< naywhayare>
this is probably a slightly better solution, at least, because it's not a requirement that the termination policy know how many iterations there are
< naywhayare>
so the user just accesses the termination policy and then uses its methods (whatever they might be) to figure out what they want to know
< naywhayare>
i.e. the ValidationRMSETermination policy might provide a LastRMSE() function, but the SimpleResidueTermination policy might provide a LastResidue() function
< sumedhghaisas>
humm... this sounds good... I already have Iteration() function I will just add MaxIteration() function to termination policy...
< sumedhghaisas>
ohh I am calling that function Index()... so termination policy can use whatever index it wants but returns that index with function Index()
< jenkins-mlpack>
andrewmw94: more tests. Really silly bug fix.
Anand has joined #mlpack
Anand has quit [Ping timeout: 246 seconds]
Anand has joined #mlpack
Anand has quit [Ping timeout: 246 seconds]
udit_s has joined #mlpack
< marcus_zoq>
naywhayare: A simple and neat solution. Thanks!
< naywhayare>
marcus_zoq: sure, no problem. anything is possible with templates... as long as you don't mind 1000-line-long error messages from the compiler :)
udit_s has quit [Quit: Leaving]
< sumedhghaisas>
naywhayare: to run a test of SVDBatchTest the command is "./mlpack_test -t SVDBatchTest/ConvergenceTest" right??
< naywhayare>
yeah
< naywhayare>
the test suite name is the first part
< naywhayare>
if you have a test suite named SVDBatchTest then you can run it all with just bin/mlpack_test -t SVDBatchTest
< sumedhghaisas>
ohh sorry I made a spelling mistake in test.cpp file.. so it was not getting recognized...
< naywhayare>
oops :)
andrewmw94 has quit [Quit: Leaving.]
sumedh_ has joined #mlpack
andrewmw94 has joined #mlpack
sumedhghaisas has quit [Ping timeout: 240 seconds]
< sumedh_>
naywhayare: regularization test for SVD incremental learning is not working...
< sumedh_>
I am unable to find regularization parameters for which RMSE is lesser...
< sumedh_>
can you just check the implementation of regularization once??
< naywhayare>
sure, I will do that, but I have a few things I am working on right now
< naywhayare>
can you check the latest files, including the non-working test, into svn?
< naywhayare>
just comment out the test that doesn't work
< sumedh_>
okay sure...
< jenkins-mlpack>
Starting build #2018 for job mlpack - svn checkin test (previous build: SUCCESS)