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: 240 seconds]
jbc__ has joined #mlpack
jbc__ has quit [Remote host closed the connection]
jbc__ has joined #mlpack
jbc__ has quit [Quit: jbc__]
sumedhghaisas has joined #mlpack
sumedhghaisas has quit [Ping timeout: 264 seconds]
< jenkins-mlpack>
Ryan Curtin: Return statements for operator++() and operator--().
< jenkins-mlpack>
Starting build #2064 for job mlpack - svn checkin test (previous build: SUCCESS)
Anand has joined #mlpack
< Anand>
Marcus : So, do you want to print the table alongside the chart? Also, what I really wanted to ask was how to generate that HTML file from the template? Some kind of "return" statements ?
< Anand>
And yes I will look into the responsiveness too.
< marcus_zoq>
Anand: I think we can print the table below the bar chart. You can iterate through all methods and as I said in the mail generate the corresponding metric.csv file. Afterwards, you can use this file and the template to create the HTML file. Open a new file (.html) and save all templates with the correct metric file. You should add a html header, does this make sense?
< Anand>
Ok. I guess we are already iterating over methods. So, I just need to save the template with the correct metrics.csv file as a string and then when the iteration is over I should write the string to an actual .html file. Right?
< marcus_zoq>
Anand: Right!
< Anand>
Alright! I am not exactly sure about how to make it responsive. But I will look into it.
< marcus_zoq>
Anand: I'm not sure, maybe d3.j salready takes care of it?
< Anand>
Marcus : I don't think it does that completely. I guess there are some options to do this. I need to explore more.
< jenkins-mlpack>
* Ryan Curtin: Fix error in random sample generation.
< jenkins-mlpack>
* Ryan Curtin: Fix out-of-order initialization warnings.
sumedhghaisas has joined #mlpack
< sumedhghaisas>
naywhayare: siddharth was going to make some changes to CF module right??
< sumedhghaisas>
I see no changes...
< sumedhghaisas>
he was going to add some SFINAE techniques...
< sumedhghaisas>
I am adding a dataset normalization techniques... and designing abstraction for them... thats why I wanted to know this...
< naywhayare>
sumedhghaisas: yeah, he has not done that yet
< naywhayare>
maybe the best idea is to go ahead and make your changes, and then later figure out how to merge them
< sumedhghaisas>
ohh... okay then... do you remember we talked about adding normalizations to dataset before factorizing them??
< sumedhghaisas>
naywhayare: you there?? I actually wanted to add those normalizations now... as now we have SVD factorizations to factorize matrices with negative entries...
< sumedhghaisas>
wanted to discuss that abstraction with you...
< naywhayare>
can you explain the normalization process?
< sumedhghaisas>
naywhayare: ohh sorry, I didn't see your msg... :(
< sumedhghaisas>
yes surely...
< sumedhghaisas>
basically I am referring to this paper...
< sumedhghaisas>
okay wait... I am looking for that paper...
< sumedhghaisas>
okay got it...
< sumedhghaisas>
Evaluation of standard SVD-based techniques for Collaborative Filtering
< sumedhghaisas>
do you have this paper??
< naywhayare>
yes, I have a copy of that
< naywhayare>
let me read it
< naywhayare>
are you talking about step 2 in section 3.1? (SVD-Cf)
< sumedhghaisas>
yes... precisely...
< sumedhghaisas>
I was thinking templatizing cleanData function...
< sumedhghaisas>
ohh yes... there should be a ticket on this...
< sumedhghaisas>
ticket #337
< sumedhghaisas>
this will close that ticket too...
< naywhayare>
but the problem is that these normalization steps don't make sense when you are using algorithms such as the ones that you have implemented, which only consider nonzero values when performing SVD
< sumedhghaisas>
ahh yes... good point to think on...
< naywhayare>
so while the normalization that you've suggested may be helpful, I don't think it will help every algorithm
< sumedhghaisas>
I think you are right...
< naywhayare>
so if you want to provide normalization, we should definitely find a way to make it optional, because it doesn't always make sense to do that
< sumedhghaisas>
its definitely does not help NMF either...
< naywhayare>
so I would think maybe a template parameter could be useful; 'NormalizationType' or something, which defaults to a policy that doesn't do anything
< sumedhghaisas>
so maybe QUICK-SVD...
< naywhayare>
Siddharth tried a similar normalization procedure for quic-svd but it didn't help results at all
< sumedhghaisas>
yes... thats what I was thinking...
< sumedhghaisas>
ooohhh... what do you suggest??
< naywhayare>
we first need to find a situation where normalization actually helps
< naywhayare>
and then we can decide what to do -- whether we should make the normalization a template parameter of CF, or of whatever decomposition algorithm it helps
< sumedhghaisas>
that would be actual SVD calculation as given in that paper...
< sumedhghaisas>
but I guess no one is going to use that...
< sumedhghaisas>
its too computationally expensive..
< sumedhghaisas>
do you think we should provide a normal SVD computation option in CF??
< naywhayare>
sure, I don't think that's a bad idea. but if we're going to do that, the class that does the normal SVD computation should do the normalization too
< sumedhghaisas>
I think its better that cleanData should that... design wise... as it is the function which does the preprocessing on the input data...
< sumedhghaisas>
so all the preprocessing should be done there...
< sumedhghaisas>
so the user can provide any type of normalization to CF class ...
< sumedhghaisas>
ohh great armadillo implements svd()
sumedhghaisas has quit [Ping timeout: 264 seconds]
sumedhghaisas has joined #mlpack
< sumedhghaisas>
naywhayare: okay some net issues... did you send me any msges in the meantime??
< naywhayare>
I didn't
< naywhayare>
so you think that CleanData() should be a member function of CF?
< naywhayare>
(where CleanData() implements the normalization in the paper you linked to)
< sumedhghaisas>
cleanData() is already a function of CF.. it takes file input and converts it into sparse matrix...
< sumedhghaisas>
so lets templatize cleanData() ... so that while converting that data into sparse matrix the template will do the normalization...
< sumedhghaisas>
what do you think??
< jenkins-mlpack>
Starting build #2065 for job mlpack - svn checkin test (previous build: SUCCESS)
sumedhghaisas has quit [Ping timeout: 264 seconds]