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/
Guest94490 has quit [Ping timeout: 244 seconds]
aminenouira has joined #mlpack
addoniis1990 has joined #mlpack
addoniis1990 has quit [Ping timeout: 250 seconds]
aminenouira has quit [Quit: Leaving]
toshad has quit [Quit: Connection closed for inactivity]
mtr_ has quit [Quit: Connection closed for inactivity]
chrishenx has quit [Ping timeout: 268 seconds]
chrishenx has joined #mlpack
agobin has quit [Quit: Connection closed for inactivity]
mkd has joined #mlpack
mkd has quit [Client Quit]
ank_95_ has joined #mlpack
ftuesca has quit [Quit: Leaving]
palashahuja has joined #mlpack
chrishenx has quit [Ping timeout: 248 seconds]
chick_ has quit [Quit: Connection closed for inactivity]
chrishenx has joined #mlpack
nilay has joined #mlpack
KeonKim has joined #mlpack
skon46 has joined #mlpack
Keon has joined #mlpack
KeonKim has quit [Ping timeout: 264 seconds]
nilay has quit [Ping timeout: 250 seconds]
Nilabhra has joined #mlpack
< Keon>
is there any performance testing tool that mlpack uses?
< palashahuja>
zoq, I have tried looking up something related to pretraining GoogLeNet, but I haven't found anything constructive
< palashahuja>
Same goes for ensemble learning in GoogLeNet
Awcrr has quit [Read error: Connection reset by peer]
Awcrr has joined #mlpack
< rcurtin>
Keon: sure, no problem. mlpack does use this benchmarking project that zoq made for GSoC in 2013: https://github.com/zoq/benchmarks/
< rcurtin>
maybe that is helpful? I don't know how helpful it would be for your project, since the benchmarking system there is generally meant to work with machine learning algorithms, not necessarily data loading/saving
< rcurtin>
but it's possible that you could use that system to get good benchmarks for what you've proposed
< zoq>
palashahuja: I guess, you can always use ILSVRC'12 to pretrain the model and e.g. ILSVRC'14 for finetuning. So I think a good idea is to pretrain the model on e.g ILSVRC'14 and to provide the parameter (weights, etc.) for other users. Does that sound reasonable?
< palashahuja>
yes, thank you that's exactly what I was looking out for
< zoq>
palashahuja: Sure no problem, I'm glad I could help.
< rcurtin>
ILSVRC, they should have picked an easier to pronounce name for their conference :)
< palashahuja>
:|
< palashahuja>
:)
< zoq>
I guess everyone just uses ImageNet (ImageNet Large Scale Visual Recognition Challenge)
< palashahuja>
:D
< palashahuja>
ImageNet sounds cooler :D
< palashahuja>
sound super-heroish ..
< rcurtin>
:)
Nilabhra has quit [Remote host closed the connection]
anatolemoreau3 has joined #mlpack
anatolemoreau2 has quit [Ping timeout: 264 seconds]
Awcrr has quit [Ping timeout: 248 seconds]
anatolemoreau3 has quit [Ping timeout: 248 seconds]
Harsh has joined #mlpack
Harsh is now known as Guest1350
Guest1350 has quit [Client Quit]
skon46 has joined #mlpack
Mathnerd314 has joined #mlpack
virtualgod has joined #mlpack
skon46 has quit [Ping timeout: 244 seconds]
awhitesong has joined #mlpack
mentekid has quit [Ping timeout: 248 seconds]
MathAddict has quit [Ping timeout: 240 seconds]
anatolemoreau3 has joined #mlpack
nilay has joined #mlpack
meenaerian has joined #mlpack
DipeshCS has joined #mlpack
< meenaerian>
Hi, I want to know more about joining google summer of code with mlpack; I'm not sure if I am qualified; am I supposed to be good at machine learning? or to be ready to learn?
< meenaerian>
Hi, I want to know more about joining google summer of code with mlpack; I'm not sure if I am qualified; am I supposed to be good at machine learning? or to be ready to learn?
< zoq>
meenaerian: 'The "necessary knowledge" sections can often be replaced with "willing to learn" for the easier projects, and for some of the more difficult problems, a full understanding of the description statement and some coding knowledge is sufficient.'
meenaerian_ has joined #mlpack
meenaerian has quit [Ping timeout: 250 seconds]
anatolemoreau4 has joined #mlpack
anatolemoreau3 has quit [Read error: Connection reset by peer]
mentekid has joined #mlpack
ftuesca has joined #mlpack
< DipeshCS>
Hello, I have submitted initial draft of proposal for "Implement Tree Types". Can any of the mentors please review and give some feedback?
meenaerian_ has quit [Ping timeout: 250 seconds]
DipeshCS_ has joined #mlpack
DipeshCS has quit [Ping timeout: 250 seconds]
DipeshCS_ has quit [Quit: Page closed]
< palashahuja>
zoq, rcurtin I have submitted my second project proposal for the GoogLeNet architecture. Please review it whenever convenient
miku_ has joined #mlpack
miku_ has quit [Client Quit]
miku_ has joined #mlpack
< miku_>
Hey. WRT the Frank-Wolfe project idea, are you *also* looking at the factorized matrix-norms, apart from the optimizations over vectors and matrices?
mentekid has quit [Ping timeout: 244 seconds]
anatolemoreau4 has quit [Ping timeout: 246 seconds]
mentekid has joined #mlpack
decltypeme has joined #mlpack
cr7 has joined #mlpack
cr7 has quit [Client Quit]
Neuron1k has joined #mlpack
< rcurtin>
miku_: yeah, it would be nice to be able to handle the specialized cases like that
< palashahuja>
zoq , sorry for the permissions issue
< palashahuja>
you could review it now
< rcurtin>
I think this probably comes down to just having a general enough FW implementation to handle that
< rcurtin>
the only one I can think of that would have a current use in mlpack is maybe the matrix trace norm... if I'm remembering correctly right now optimization for an SDP is over the trace norm of the input matrix
< miku_>
rcurtin: Alright, but irrespective, I'm thinking it makes more sense to focus on the optimization routines. You'd agree?
< miku_>
It would seem that they are the ones more relevant for ML tasks...
< rcurtin>
miku_: yes, I agree, we should focus on the implementations that are directly applicable to the types of stuff mlpack already implements
mentekid has quit [Remote host closed the connection]
< rcurtin>
ideally though we should try and make the implementation flexible, so that people can easily implement future support for other optimization domains
Bartek has joined #mlpack
< miku_>
Sure, will make sure to allow that.
< miku_>
From the ideas page, it seemed that you guys thought the OMP was one of the big things mlpack could take away from a Frank-Wolfe implementation. Is there any other use case you guys are particularly looking at?
< rcurtin>
I didn't write the idea; I think Stephen was very interested in OMP
< rcurtin>
personally I am interested in having it as an available optimizers to complement the existing optimizers
nilay has quit [Ping timeout: 250 seconds]
ank_95_ has quit [Quit: Connection closed for inactivity]
< decltypeme>
rcurtin A question For GSoC proposal, what are the suggested datasets to test against (for classification), like the famous iris dataset for example? What else?
< rcurtin>
hmmm, there's a lot of stuff on the UCI dataset repository that I usually do, but they are often very big datasets
< rcurtin>
for a unit test it's probably best to stick with something little like iris, or maybe some simple synthetic data you've generated
< rcurtin>
but if you are looking for larger datasets for classification, you might try covertype, pokerhand, connect4, mnist
< rcurtin>
I think higgs and susy are classification too, but they are extremely large (close to 10M points in 10-20 dimensions)
kirizaki has joined #mlpack
< miku_>
Any heads-up for testing the FW algorithms?
< miku_>
Was wondering what sort of tasks this will work best for, classification or regression...
< rcurtin>
I actually don't have the knowledge to be sure which tasks the FW optimizer will be best for
< rcurtin>
for testing, we should basically just make sure that it is able to converge to reasonable solutions
< rcurtin>
you can take a look at how the other optimizers are tested
< rcurtin>
like src/mlpack/tests/sgd_test.cpp, and for instance some of the tests in src/mlpack/tests/logistic_regression_test.cpp
tsathoggua has quit [Quit: Konversation terminated!]
virtualgod has quit [Ping timeout: 240 seconds]
kirizaki has quit [Remote host closed the connection]
virtualgod has joined #mlpack
agobin has joined #mlpack
< toshad>
rcurtin: I have shared my proposal on the GSoC website. It would be great if you could review it and suggest any improvements in your free time
awhitesong has joined #mlpack
mrbean has quit [Read error: Connection reset by peer]
mrbean has joined #mlpack
< miku_>
rcurtin: I've shared my proposal on the site, would really like to get feedback from you. Thanks.
addonis1990 has joined #mlpack
nataliat95 has joined #mlpack
miku_ has quit [Ping timeout: 250 seconds]
miku_ has joined #mlpack
KeonKim has joined #mlpack
Keon has quit [Ping timeout: 252 seconds]
miku_ has quit [Ping timeout: 250 seconds]
addonis1990 has quit [Remote host closed the connection]