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/
mikeling has joined #mlpack
govg has quit [Ping timeout: 260 seconds]
govg has joined #mlpack
vinayakvivek has joined #mlpack
travis-ci has joined #mlpack
< travis-ci> dineshraj01/mlpack#4 (rsmodel-boost-variant-visitor - 5580af2 : dinesh Raj): The build is still failing.
travis-ci has left #mlpack []
dinesh_ has joined #mlpack
dinesh_ has quit [Ping timeout: 256 seconds]
dinesh_ has joined #mlpack
dinesh_ has quit [Remote host closed the connection]
dinesh_ has joined #mlpack
dinesh_ has quit [Ping timeout: 255 seconds]
dinesh_ has joined #mlpack
govg has quit [Ping timeout: 255 seconds]
dinesh__ has joined #mlpack
dinesh_ has quit [Ping timeout: 255 seconds]
dinesh__ has quit [Ping timeout: 255 seconds]
dinesh_ has joined #mlpack
dinesh_ has quit [Remote host closed the connection]
dinesh_ has joined #mlpack
sumedhghaisas has joined #mlpack
dinesh_ has quit [Quit: Leaving]
< vivekp> Hi, I'm facing a problem with build process.
< vivekp> it appears that make doesn't build ann module along with a bunch of others e.g. amf, approx_kfn etc.
< vivekp> I'm working on top of latest git master.
< vivekp> Did I miss something ?
< rcurtin> vivekp: what's the command you're using to build?
< vivekp> make after configuring cmake inside build directory
< rcurtin> what happens when you type 'make mlpack_approx_kfn'?
< vivekp> okay, let me check
< vivekp> it starts building that module
< rcurtin> and if you type make it doesn't?
< vivekp> and with make_ann, I get an error: make: *** No rule to make target 'mlpack_ann'.
< vivekp> rcurtin: yes
< rcurtin> so if you type make, it never builds mlpack_approx_kfn?
< rcurtin> and you have not modified the CMake configuration?
< vivekp> make mlpack_ann* -- sorry about that
< rcurtin> there is no target mlpack_ann, that does not exist
< rcurtin> mlpack is made up of a core library ('make mlpack'), header files ('make mlpack_headers'), and command-line programs ('make mlpack_approx_kfn', 'make mlpack_knn', etc...)
< rcurtin> so there are some cases where a command-line program does not exist for a particular set of code in mlpack/methods/
< vivekp> rcurtin: okay, understand
< vivekp> and sorry about the confusion. I just had a second look and the modules that are never built are amf and ann
< rcurtin> no worries---the AMF code does not actually get built into a command-line program either
< rcurtin> that is used in mlpack_nmf and mlpack_cf, but there is no mlpack_amf program
< vivekp> ok, understood. Thank you for your prompt reply
< rcurtin> sure, sorry if it was a little short, I am dealing with a bunch of things at once
< rcurtin> time-division multiplexing my responses causes the quality to suffer a bit :)
< vivekp> not a problem at all. I got my doubt resolved :)
< rcurtin> sure, glad I could help---feel free to ask any more questions if needed
< rcurtin> I've found hosting for a system to add to our build server; it has 64 cores and 192GB RAM, so I hope it will come online in the next few weeks
< rcurtin> there are still a few details to work out (including what it will be named, of course), but this will add to our already-overpowered build server setup :)
< rcurtin> and when I figure out what the issue is with the three 64-core sparc64 machines is (they hang, it must be a kernel bug or something), then we'll have an even more overpowered setup
< zoq> wow, lying around somewhere? The build setup is going to be awesome.
< sumedhghaisas> zoq: Hey Marcus
< sumedhghaisas> Actually I will like to work on it as a student
< rcurtin> zoq: yeah, it was originally a system that was built for the FASTLab, but then the FASTLab kind of disbanded while I was in graduate school and nobody was left but me to take care of all the equipment
< rcurtin> so I had to spend some time negotiating with connections I still had on campus to see if someone was willing to sponsor it, etc., and finally I heard back and the answer is yes, so hopefully it will come online soon
< zoq> sumedhghais: ah okay, great. So yes, I think some MANN like NTM + merge + batch norm should be enough work we could also think about adding some other features like recurrent dropout.
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#1781 (master - 28202d7 : Ryan Curtin): The build passed.
travis-ci has left #mlpack []
< zoq> rcurtin: Sounds great, same location as for the benchmarks/masterblaster machines?
< sumedhghaisas> zoq: okay. I will read about NTM and get back to you. I am confident about merge and batch normalization but I have to read about NTM a little
< zoq> sumedhghais: Sounds good, you could also take a look at other MANN's if you like, it hasn't to be NTM.
< sumedhghaisas> zoq: can you please refer me some papers that I can look at?
< rcurtin> zoq: nope, this system will be hosted in atlanta; masterblaster is in Los Angeles, and the benchmarking systems are in Oregon. looks like masterblaster will have to move to Oregon, so I'll have to get that process in motion
< zoq> sumedhghais: I would recommend to take a look at e.g. the DNC (Hybrid computing using a neural network with dynamic external memory) a successor of the NTM model or the HAM (Hierarchical Attentive Memory) paper, both papers have a lot of references to other ideas.
< zoq> rcurtin: I guess, since you know the atlanta guys no hassle with opening ports etc.
< rcurtin> yeah, this system will be located on the Georgia Tech campus, and I still know all the support people and their procedures, so no stupid problems with ports
< rcurtin> Symantec IT will not be involved at all :)
< zoq> I guess, the atlanta machine doesn't have a Nvidia GPU, just asking :)
< sumedhghaisas> zoq: Sure. I will take a look.
< rcurtin> unfortunately no, but I am still fighting to get some GPUs for either masterblaster or other systems
< rcurtin> everything moves glacially slow at Symantec, so I thought we were ordering new GPUs three months ago, but it still hasn't happened
< zoq> Maybe one day :)
< rcurtin> yeah, we'll see...
< rcurtin> part of the problem is that most of the company is out in mountain view and I'm here in atlanta, so I can't regularly ask the guy on our team who orders GPUs what the status is
< rcurtin> I could send him emails every day but that might be a bit more rude and annoying than if I was asking him in person every few days :)
< zoq> Absolutely, I'm confident that one day someone says "Hey, here is the Graphics Card you asked for".
< rcurtin> one day in 2019 maybe :)
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#1782 (master - 6b2d239 : Ryan Curtin): The build was broken.
travis-ci has left #mlpack []
mikeling has quit [Quit: Connection closed for inactivity]
benchmark has joined #mlpack
benchmark has quit [Client Quit]
ashk43712 has joined #mlpack
ashk43712 has quit [Ping timeout: 260 seconds]
travis-ci has joined #mlpack
< travis-ci> mlpack/mlpack#1784 (master - 5daaea2 : Ryan Curtin): The build is still failing.
travis-ci has left #mlpack []
ashk43712 has joined #mlpack
sumedhghaisas has quit [Ping timeout: 245 seconds]
< ashk43712> Hi, I am Ashish. I am new to mlpack and want to contribute. I have compiled and installed mlpack. I found the idea of "Alternatives to neighborhood-based collaborative filtering" very interesting for GSOC17. I have a good C++ and Machine Learning knoeledge. Can someone point me to the right direction so that I can start contributing by solving some bugs or adding new features
< rcurtin> ashk43712: hi there, have you seen http://www.mlpack.org/involved.html or http://www.mlpack.org/gsoc.html?
< rcurtin> I know there are not many bugs open for the CF package right now, but actually I was just talking to someone yesterday and thought of a good one...
< rcurtin> if you run mlpack_cf on the movielens dataset with the RegSVD optimizer, it performs well and gets like 0.8 RMSE
< rcurtin> but if you use NMF or SVDCompleteIncremental or SVDIncompleteIncremental or SVDBatch, then the RMSE is very bad
< rcurtin> I think this is because the default parameters for those optimizers are not good
< rcurtin> so if you wanted, you could see if you can reproduce that issue, and then see if you can find parameters for the optimizers that give better results
< rcurtin> I know that maybe that is a lot to ask as a first contribution but it's what I can think of right now :)\
< ashk43712> okay, thanks a lot
benchmark has joined #mlpack
benchmark has quit [Client Quit]
< rcurtin> good to know my addition of a space to the range search code did not affect the runtime :)
< zoq> :)
ashk43712 has quit [Quit: Page closed]
vinayakvivek has quit [Quit: Connection closed for inactivity]