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/
< jenkins-mlpack>
* Ryan Curtin: Remove HasParentDistance trait, becase it wasn't used anywhere.
< jenkins-mlpack>
* andrewmw94: now add the actual files the way they are supposed to exist. No real code yet.
< jenkins-mlpack>
* andrewmw94: remove the silly folders I added when I meant to add txt files.
< jenkins-mlpack>
* sumedhghaisas: added module 'lmf'(Latent Matrix Factorization) to accommodate SVD based update rules alongside NMF based update rule. CF module is updated to use LMF module.
< jenkins-mlpack>
* andrewmw94: added more code for the rectangle type trees
< jenkins-mlpack>
* Ryan Curtin: Remove the Pelleg-Moore k-means implementation; it is being replaced.
witness___ has joined #mlpack
Anand has joined #mlpack
Anand has quit [Ping timeout: 240 seconds]
udit_s_ has joined #mlpack
udit_s_ has quit [Client Quit]
udit_s has joined #mlpack
< marcus_zoq>
Hello udit, I will look at your code in a few minutes. The next step would be to write tests for the DecisionStump class.
andrewmw94 has joined #mlpack
< udit_s>
okay.
sumedhghaisas has joined #mlpack
sumedhghaisas has quit [Ping timeout: 264 seconds]
sumedhghaisas has joined #mlpack
sumedhghaisas has quit [Ping timeout: 240 seconds]
witness___ has quit [Quit: Connection closed for inactivity]
andrewmw94 has quit [Quit: andrewmw94]
andrewmw94 has joined #mlpack
< naywhayare>
udit_s: can I get a link to a reference describing the maximum-entropy splitting algorithm you are using?
andrewmw94 has left #mlpack []
andrewmw94 has joined #mlpack
udit_s has quit [Quit: Ex-Chat]
< jenkins-mlpack>
Starting build #1913 for job mlpack - svn checkin test (previous build: SUCCESS)
< naywhayare>
andrewmw94: last night I removed the TreeTraits<>::HasParentDistance property; it wasn't used anywhere
< andrewmw94>
yeah, I saw that
< andrewmw94>
I added a comment to the baseCase() function. I don't know if there was a reason it didn't already have a description?
< andrewmw94>
I'll commit so you can see it
< naywhayare>
yeah, thanks; probably the reason it didn't have a comment was that it was overlooked
< andrewmw94>
also, it might be nice to add a comment somewhere in the trees explaining that the RuleType stuff is in the methods/neighborSearch folder. But that's more for understanding the code than for the typical end user
< andrewmw94>
and I'm not really sure where such a comment would go
< naywhayare>
I'm planning to write a tutorial on mlpack trees and dual-tree algorithms at some point
< andrewmw94>
ahh
< naywhayare>
and that would probably explain things well, but I have a paper deadline coming up in early June so it has to wait at least until then...
< andrewmw94>
yeah. Deadlines make it so nothing ever gets done
< jenkins-mlpack>
* Ryan Curtin: There's nothing in this directory and it appears to be unused.
< jenkins-mlpack>
* Ryan Curtin: The system hotwheels.cc.gt.atl.ga.us doesn't exist anymore.
Anand has quit [Ping timeout: 240 seconds]
koderok has joined #mlpack
koderok has quit [Client Quit]
koderok has joined #mlpack
Anand has joined #mlpack
< Anand>
Hi Marcus!
< Anand>
So, you were talking about the quadratic loss function. Right?
< marcus_zoq>
Hi Anand, right.
< Anand>
It takes into account both the probabilities and the actual class
< Anand>
We deal with two vectors for each instance
koderok has quit [Ping timeout: 252 seconds]
< marcus_zoq>
Yeah, thats correct.
< Anand>
Seems a nice way to implement the metric
< Anand>
But do we have the way probabilities?
< Anand>
*the probabilities
< marcus_zoq>
Yeah, if we modify the nbc method.
< Anand>
Ok. I will have a look into it
< marcus_zoq>
It's just a small modification of the nbc_main.cpp file.
< Anand>
In nbc_main.cpp?
< Anand>
How?
< marcus_zoq>
Maybe it's not just a small modification, let's check :)
< Anand>
We will need to generate another file other than output.csv wherein each line will contain the probabilities of an instance being in each class
koderok has joined #mlpack
< marcus_zoq>
That would be good. Okay the propabilities are stored in pointProbs (line 144 naive_bayes_classifier_impl.hpp).
< Anand>
Did you just add that?
< Anand>
I need to build mlpack again then!
< marcus_zoq>
No, we need to find a proper way to store the values :)
< Anand>
So what did you do?
< Anand>
I donot have pointProbs in my file
< Anand>
:P
< marcus_zoq>
I'm talking about this line: 'arma::vec pointProbs = testProbs.row(i).t();'
< Anand>
In src/mlpack/methods/naive_bayes_classifier_impl.hpp?
< Anand>
It is not there in mlpack 1.0.8 that I downloaded from mlpack.org
< marcus_zoq>
Oh, I use the bleeding edge version :)
< Anand>
So, I need to check out from svn then and build again?
< marcus_zoq>
Yeah!
< Anand>
Man!
< Anand>
Please update the mlpack page too! :P
< Anand>
Ok. After that. the pointProbs vector needs to be written to a file
< marcus_zoq>
naywhayare: We need the probability for each of the data points. I think the simplest way to do that is to getter that returns arma::mat testProbs?
< jenkins-mlpack>
* andrewmw94: add a short and hopefully useful explanation of what different files are for, focusing on BSP trees.
< jenkins-mlpack>
* andrewmw94: added comment to base case. Some more preliminaries for rectangle trees.
< naywhayare>
andrewmw94: another class that's coded separately, in addition to bounds, metrics, splits, rules, and traversals, are statistics
< naywhayare>
these statistics are generally used to cache information about the nodes that are relevant to certain tasks
< naywhayare>
for instance, in nearest neighbor search, the NeighborSearchStat class is used as StatisticType, and it keeps track of the current bound for a query node
< naywhayare>
where the current bound is that complicated B(N_q) function from the tree-independent dual-tree algorithms paper (or something similar)
koderok has quit [Quit: koderok]
andrewmw94 has quit [Ping timeout: 276 seconds]
andrewmw94 has joined #mlpack
< andrewmw94>
sorry about that. I hate my internet connection
< naywhayare>
it's comcastic!
< andrewmw94>
I'm confused on the mrkd_statistic_impl.hpp
< andrewmw94>
it seems like most of it is accidentally commented out
< andrewmw94>
so I assume that the file isn't really used?
< naywhayare>
for some reason the routes from my house to campus, which is less than half a mile away, give all kinds of SSH lag. I'll never know why, because I'll never be able to talk to anyone at Comcast who actually knows anything about anything
< naywhayare>
yeah, MRKDStatistic was for a kind of k-means clustering with trees, but it was written only for binary space trees
< naywhayare>
I kept the file around because "I'll fix it someday"
< naywhayare>
well, the day I'm fixing it actually happens to be today, but I'm going to throw MRKDStatistic away because I can perform the algorithm without needing a statistic in the tree
< andrewmw94>
ahh
< naywhayare>
I had thought what I would do was take MRKDStatistic and refactor it to work with any type of tree, and then the code in KMeans::FastCluster() would work... but I think it was definitely easier to just start over and do it differently
< jenkins-mlpack>
Starting build #1915 for job mlpack - svn checkin test (previous build: SUCCESS)