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/
vivekp has quit [Remote host closed the connection]
vivekp has joined #mlpack
Arkadiy has joined #mlpack
< Arkadiy> Hi everybody! I am interested in GSOC-2017 projects. Is it still possible to find a mentor for them? Thanks
chvsp has quit [Quit: Page closed]
< zoq> Arkadiy: Hello there, do you mean projects not listed on the ideas page?
mikeling has joined #mlpack
< Arkadiy> No, projects already listed. Mostly I am interested in Reinforcement Learning project API for cross-validation and parameters tuning, because I have experience with such tasks (Of course I would be happy to tell much more about me if it still possible to apply to them:) )
< Arkadiy> (I mean 2 different projects surely)
< zoq> Arkadiy: Sure, you can apply until the student application deadline. Make sure to search through the mailing list archive (http://mlpack.org/pipermail/mlpack/), there should be some pointers for each project.
< zoq> Arkadiy: Also, if you like you can introduce yourself on the mailing list.
< Arkadiy> Ok, thanks!
Arkadiy has quit [Ping timeout: 260 seconds]
trapz has joined #mlpack
govg has quit [Ping timeout: 240 seconds]
govg has joined #mlpack
kris2 has joined #mlpack
trapz has quit [Quit: trapz]
pranjal has joined #mlpack
pranjal has quit [Ping timeout: 260 seconds]
vinayakvivek has joined #mlpack
kris2 has left #mlpack []
pvskand has joined #mlpack
< pvskand> Is there svm support in mlpack?
sagarbhathwar has joined #mlpack
ironstark has quit [Ping timeout: 240 seconds]
ironstark has joined #mlpack
< sagarbhathwar> For GSoC, is implementation of affinity propogation a good idea for proposal?
sagarbhathwar has quit [Ping timeout: 260 seconds]
kris1 has joined #mlpack
kris1 has left #mlpack []
jenkins-mlpack has quit [Ping timeout: 240 seconds]
thyrix has joined #mlpack
jenkins-mlpack has joined #mlpack
jenkins-mlpack has quit [Ping timeout: 240 seconds]
jenkins-mlpack has joined #mlpack
trapz has joined #mlpack
trapz has quit [Quit: trapz]
arcady has joined #mlpack
< arcady> Hi again! Can anybody please review whether my Proof of Enrollment for GSOC is correct? Thanks!
vivekp has quit [Ping timeout: 260 seconds]
vivekp has joined #mlpack
Trion has joined #mlpack
< rcurtin> arcady: unfortunately I think it's the GSoC admins that care about that most, we don't see that at all
< rcurtin> you might consider asking in #gsoc
< rcurtin> pvskand: no, there is no svm support at this time
arcady has left #mlpack []
< rcurtin> sagarbhathwar: if you are clear about the API you will implement and how it will fit in with the rest of mlpack, then I think it's fine to propose affinity propagation
< pvskand> rcurtin : Is there a specific reason as to why there isn't the support for svm?
< Trion> Is there a way to flatten the tensor that is received after the convolution layer?
< zoq> Trion: Maybe vectorise is helpful?
< Trion> How to find class and function definitions fast? :P
< rcurtin> pvskand: nobody has implemented it; if you wanted to implement it, this is fine, but we must ensure that it's faster than libsvm and other implementations, otherwise there is no reason for anyone to use it
< pvskand> Okay! I see
thyrix has quit [Quit: Page closed]
< zoq> rcurtin: cherry-pick?
< Trion> zoq: yes vectorise is what I need, is it possible to fit it in the model API? like after using model.Add(Convolution<>(10, 15, 5, 5)) I can add it as a rule for model?
< zoq> Trion: The conv layer should return arma::mat(elem, 1) isn't that what you need, I mean it's basically arma::vec.
vinayakvivek has quit [Quit: Connection closed for inactivity]
< rcurtin> zoq: yeah, the process is somewhat intense...
< rcurtin> I have to get a list of all commits/merges to master only
diehumblex has joined #mlpack
< rcurtin> which I can do with git log --first-parent master
< rcurtin> then once I have the list of all commits directly to master since 2.1.1 (some will be direct commits to master, some are PR merges), I can use git show-branch to get the names of the branches then use git log --first-parent to get only the commits on that branch
< rcurtin> then cherry-pick them into mlpack-2.2.x
< rcurtin> maybe there is a better way, but I don't know one
< Trion> zoq: shouldn't conv layer return a [width, height, depth] cube given an image of [width, height, 3]? Is flattening being done automatically? :P that is convenient!
vss has joined #mlpack
< zoq> rcurtin: I don't envy your task ... I was thinking maybe you could make the commands you used for a new release public, that way maybe someone else can take a look and probably add some notes to improve the process.
< zoq> Trion: It is isn't it, the problem when returning something like an arma::cube is that you have to implement every layer for arma::cube and arma::mat.
< rcurtin> zoq: yeah, I was going to post on the wiki because I will probably forget by the next release :)
< zoq> rcurtin: great!
< rcurtin> it isn't very hard when there is not much time between releases... then it is only a few cherry picks
< rcurtin> but in this case it was 3 months with a huge amount of contributions :)
< Trion> But the problem arises that I am trying to do conv layer after conv layer, premature flattening will destroy the data's structure :(
< zoq> Trion: The conv layer takes care of the input shape, no worries.
< vss> watcha guys talking about ? XD
< Trion> :O wow!
< zoq> rcurtin: So release more often?
< zoq> vss: http://www.mlpack.org/irc/ feel free to jump in :)
< Trion> Trying to create sentinel agents to save the earth from aliens
< vss> :o
< zoq> :)
< vss> sentinel agents like the one from matrix ? :D
< Trion> Don't worry they can only press shoot, left and right :P they won't attack us back
< vss> oh right
< vss> you are gonna work on reinforcement training right ?
< Trion> I hope :P
< zoq> Btw. If anyone has some music recommendations I'm eager to hear it :)
< Trion> Nuttin but Stringz's Thunder is top of my playlist
< vss> what kind of music do you like ?
< Trion> just for now... >:-)
< zoq> interesting playing it right now
< Trion> I like classical(piano and violin) and edm
< zoq> reminds me of you lindsey stirling .. without the rap parts
< rcurtin> zoq: yeah, definitely more releases would be nice, hard to find the time to do that :)
< rcurtin> I think ideally it would be really cool to have an "automatic" release pipeline, but that is really time-consuming to set up
< rcurtin> I'm listening to Third Eye Blind this morning... some nice nostalgia
< rcurtin> I used to listen to this album when I was growing up, when I played Commander Keen 4 on my dad's P2 machine
< rcurtin> I'm not sure it's very good for much other than nostalgia value :)
< vivekp> zoq: Elements by Lindsey Stirling is my favorite. :)
< vss> Trion : Try listening to this https://www.youtube.com/watch?v=K5KAc5CoCuk
< vss> Lindsey Stirling is badass
< zoq> rcurtin: Interesting idea, but probably hard to realize, if you like to do it the "right" way.
< rcurtin> yeah, I agree, I think it is more work than it's worth for now
< zoq> vss: wow a lot of views
< rcurtin> if we can finish the things for mlpack 3.0.0 then it won't be necessary to maintain this extra mlpack-2.2.x branch which takes a lot of time
< rcurtin> but I still have a lot of bindings work before that is ready
< Trion> vss: :P I couldn't understand lyrics, but liked the voice
< Trion> Lindsey Stirling's music does sound similar :D Gotta catch 'em all!
< zoq> kinda differernt https://www.youtube.com/watch?v=JWnX41TBFF4 one of my favorites :)
< vivekp> Talking about views, I wonder if anyone has seen the count on this https://www.youtube.com/watch?v=9bZkp7q19f0 :)
< vivekp> it is at the top on YT in terms on views
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2109 (mlpack-2.2.x - 682e938 : Ryan Curtin): The build was broken.
travis-ci has left #mlpack []
< zoq> vivekp: that's insane
< Trion> :D It is nice! Moderat music is ambient, helping me focus
< rcurtin> yeah, I like the Moderat, I think I will add that to my playlist and check out some more
< Trion> vivekp: I didn't know people still watch Gangnam style vid, I thought the trend died :O
< vivekp> Just finished watching a couple of Moderat videos; time to add them in my playlist now!
< vivekp> trion: i can tell by looking at the comment section, very many people still do :)
< Trion> :P "Like if you are watching in 2017"?
< vivekp> Haha, yes
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2110 (mlpack-2.2.x - 5a47b85 : Lakshya Agrawal): The build passed.
travis-ci has left #mlpack []
< vss> btw there are 5 proposals that you can submit for gsoc right ? So can i submit two proposals for the same organization ( different ideas) ?
yaing has joined #mlpack
< yaing> Hello guys, i have finally setup my development environment for mlpack
< Trion> Yes you can
< yaing> i would like to work on issue https://github.com/mlpack/mlpack/issues/919
< yaing> can i start working on it if no one else is working on it?
< zoq> yaing: Nax is already working on the issue.
< yaing> okay thank you zoq
< yaing> i will pick some other issue
< vss> its harder to build the source code on windows isn't it ?
< yaing> i don't know about that but i think linux is more convenient
< vss> i guess so
< Trion> I miss my Internet Explorer :(
< yaing> which browser do you use Trion?
< Trion> chrome-dev
< yaing> nice choice Trion. zoq can you tell me about an easy issue to work on?
< zoq> yaing: Unfortunately, I can't, search through the issues on github, maybe you find something interesting you can also search through the codebase and maybe you find something that could be improved?
< yaing> okay thank you zoq. i will find something to work on
< Trion> SGD takes DecomposableFunction as argument, how to make one for ann model?
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2111 (mlpack-2.2.x - 2c4052d : Ryan Curtin): The build was fixed.
travis-ci has left #mlpack []
< yaing> i like three project ideas reinforcement learning, deep learning modules and augmented rnn. zoq can you give me some advice on each project idea so that i can write a strong proposal?
< yaing> i will submit my proposal on one of them but i have not decided which one yet
< zoq> Trion: As long as the given function implements the necessary functions like Parameters, Gradient and Evaluate you can use the SGD optimizer, so FFN implenets the mentioned functions so SGD<decltype(model)> sgd(mode, ...) take a look at src/mlpack/tests/feedforward_network_test.cpp line 69.
kris2 has joined #mlpack
< kris2> how can i make appvoyer use armadillo c++11
< kris2> ARMA_USE_CXX11
< zoq> yaing: I'm not sure I can help you with the decision since it depends on your interests and capabilities, each project has been discussed on the mailing list, so my recommendation is to go through the archive: http://mlpack.org/pipermail/mlpack/
< kris2> Currently app voyer is failing on the use initializer list
< yaing> okay, i will go through the archive. Thank you
< yaing> also, can i ask who will mentor these projects? and how can i get into contact with them?
< zoq> kris2: appveyor uses c++11, but the compiler is sometimes a little bit strange when it comes to implementation of the standard.
< kris2> okay then i would use c++98 type intialisation
< zoq> yaing: Please use the public mailing list to discuss project details, that way everyone can make comments.
< zoq> kris2: Okay, sounds good.
< kris2> one more thing i asked you about the named ffn. Should i go ahead with it or just close the PR(if it not that useful a feature).
Trion has quit [Quit: Have to go, see ya!]
< zoq> kris2: I'll go and comment on the PR later today.
< kris2> okay thanks. Also gaussian init pr i made the changes you asked and added the test so i think its ready for review. whenever you have the time please go through it.
< yaing> okay zoq i got your point. thank you again
< yaing> and i have decided to pick test failure issue
< yaing> i will start working on it now
< zoq> yaing: Sounds great!
yaing has quit [Quit: Page closed]
kartik__ has joined #mlpack
< pvskand> is mlp implemented by using the perceptron class?
kartik__ has quit [Ping timeout: 260 seconds]
nish21 has joined #mlpack
ironstark has quit [Remote host closed the connection]
vss has quit [Quit: Page closed]
Trion has joined #mlpack
Trion has quit [Ping timeout: 260 seconds]
Trion has joined #mlpack
sahilc has joined #mlpack
< sahilc> hey, I need a little help with an issue
< sahilc> is anyone free to discuss?
< rcurtin> sahilc: what is the problem?
< rcurtin> pvskand: I don't believe so
< sahilc> I was working on issue #356, and needed advice
< pvskand> What is the reason behind having single layer perceptron and ann separately?
< rcurtin> sahilc: I am not sure exactly how to help... can you ask a clear question I can help with?
< rcurtin> pvskand: I think the reason is that Udit implemented the perceptron, and then later Marcus did the ANN code :)
< rcurtin> I think it would be ideal to reimplement the perceptron as part of the ANN package
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2115 (mlpack-2.2.x - b7e60e9 : Ryan Curtin): The build was broken.
travis-ci has left #mlpack []
mandroid6 has joined #mlpack
< sahilc> I have replicated the emst code to build a MST. Now, I need to modify this to fit single linkage use case. Ideally, a user specifies a parameter for cluster numbers where we stop merging clusters. Also, we should replicate nearest neighbors output for this clustering method?
mandroid6 has quit [Ping timeout: 260 seconds]
< rcurtin> sahilc: yes, I think it would be reasonable if the user specified a number of clusters that they wanted
< rcurtin> I'm not sure what you mean by 'nearest neighbors output'
< sahilc> mlpack has a "nearest neighbors" module which dumps the final clusters into a file. We could replicate that for hierarchical clustering
mikeling has quit [Quit: Connection closed for inactivity]
< rcurtin> sahilc: it doesn't dump the final clusters into a file, it dumps the nearest neighbors into a file
< rcurtin> but I agree, it would be useful to output both the final cluster centroids and the assignments of each point to a centroid into a file
< rcurtin> see the k-means program for an idea
< sahilc> Thanks! Will send a PR soon :)
< pvskand> rcurtin : I was thinking of implementing dropout for fnn. Is it okay? or has it been implemented?
< pvskand> Oh my bad! Didn't see the dropout file! Sorry about that
Trion has quit [Quit: Have to go, see ya!]
pvskand has quit [Ping timeout: 260 seconds]
pvskand has joined #mlpack
pvskand has quit [Ping timeout: 260 seconds]
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2118 (mlpack-2.2.x - e4551bf : Ryan Curtin): The build is still failing.
travis-ci has left #mlpack []
pvskand has joined #mlpack
< pvskand> zoq : I was planning to implement mlp_forward and mlp_backward in MATLAB in the benchmarks. Is that fine?
< rcurtin> pvskand: you already opened an issue for this, I am sure that you will get an answer when someone has a chance; no need to ask multiple times
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2121 (mlpack-2.2.0 - d9164ab : Ryan Curtin): The build has errored.
travis-ci has left #mlpack []
pvskand has quit [Ping timeout: 260 seconds]
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2123 (mlpack-2.2.0 - 95867f7 : Ryan Curtin): The build has errored.
travis-ci has left #mlpack []
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2122 (mlpack-2.2.x - 95867f7 : Ryan Curtin): The build has errored.
travis-ci has left #mlpack []
jenkins-mlpack has quit [Remote host closed the connection]
jenkins-mlpack has joined #mlpack
nish21 has quit [Ping timeout: 260 seconds]
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2124 (mlpack-2.2.x - 40bf682 : Ryan Curtin): The build has errored.
travis-ci has left #mlpack []
< rcurtin> whew, I think that is the most time I've ever had to put into an mlpack release
< rcurtin> I think that was basically 4 straight days, although a lot of that was fixing problems before the release process could start
< rcurtin> I'm going to go take care of step 22 on the release preparation guidelines now
< zoq> Maybe the reason why there is something like step 22 is because of this tedious process.
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#2126 (master - 81dc1e5 : Ryan Curtin): The build passed.
travis-ci has left #mlpack []
sahilc has quit [Quit: Page closed]
s1998 has joined #mlpack
jenkins-mlpack has quit [Remote host closed the connection]
jenkins-mlpack has joined #mlpack
< kris2> zoq: if you have time today could you please review Xavier Init and Gaussian Init today. As i have a free day tommrow. So i could make the changes. Thanks.