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/
travis-ci has joined #mlpack
< travis-ci>
mlpack/mlpack#1544 (master - 045c16a : Ryan Curtin): The build passed.
< rcurtin>
weird, seems like it ran the neighbor search benchmarks but the only file I changed was lsh_test.cpp
sumedhghaisas has joined #mlpack
< zoq>
I've hardcoded the neighbor search benchmark to test the output.
< rcurtin>
ah, ok, well that would explain it :)
< rcurtin>
the Sun systems are being set up right now, but the person working on them has not managed to get the serial console set up right yet
< rcurtin>
so, soon, we will be able to reliably test mlpack on sparc64 :)
< rcurtin>
I wonder if Docker will even run on sparc64, so that I can use docker images for different configurations
< zoq>
I've always wanted to build mlpack on sparc :)
< rcurtin>
oh sweet! the CoreOS rkt project says that it builds on multiple architectures, so maybe I can build that for sparc64 and use that as a replacement for docker
< rcurtin>
I think the rkt maintainers will be confused if I start filing bugs about sparc64, so hopefully I can find some :)
< zoq>
:) the whole build is going to be awesome
< zoq>
I don't think a lot of projects could say, hey we make sure everything builds on sparc64 :)
< rcurtin>
haha
< rcurtin>
I have an ARM chromebook I'll hook up as a permanent build slave too, because I never seem to actually use the system
< zoq>
:), I don't think we could use the chromebook to build against nvblas?
< rcurtin>
I doubt it, I don't think it has a GPU that's worth anything
< rcurtin>
I've sent an email to someone I know inside the company to see if there is a spare GPU anywhere we could throw into masterblaster
< rcurtin>
I am sure it will probably take weeks to put it in there and I will have to fill out forms :)
< zoq>
oh for sure, but it would be a great addition
sumedhghaisas_ has joined #mlpack
sumedhghaisas has quit [Read error: Connection reset by peer]
sumedhghaisas_ has quit [Ping timeout: 244 seconds]
< rcurtin>
zoq: I'll be giving a talk on mlpack at a conference called MLconf this Friday, and I'd like to mention the experimental deep learning support, so it might be possible that it will garner some interest afterwards
< rcurtin>
I'll also get a chance to mention all the great GSoC projects too, so hopefully the talk will gather some interested users and contributors
< zoq>
rcurtin: sounds great, I'm about to about some things which should be ready by the end of the next week :)
< zoq>
Also mentioning all GSoc projects is also a terrific idea.
< rcurtin>
I did have a question though, is the intention that the entire network structure is known at compile time and therefore can be optimized at compile time?
< rcurtin>
kind of like an armadillo expression?
< rcurtin>
I did some digging, and it seems Caffe, TensorFlow, and other C++ libraries for deep learning don't do that---they load the model structure at runtime
< rcurtin>
I think maybe dlib-ml does it at compile time but I have not dug into it very far
sumedhghaisas_ has joined #mlpack
sumedhghaisas_ has quit [Read error: Connection reset by peer]
< zoq>
Right, right now the entire network structure (all layer) has to be known at compile time.
< zoq>
Another thing, since we have all these nice machines I was thinking it would be great to run more than one benchmark job at the same time. Unfortunately, sqlite isn't the best solution to handle multiple writes from different hosts. So I was thinking it might be a good idea to use mysql instead.
< rcurtin>
hm, yeah, that might be a good solution
< rcurtin>
we can run mysql on masterblaster or I can try to find a different host
< rcurtin>
but it seems like we shouldn't run an extra process on the benchmark hosts if possible
< zoq>
I agree, probably the machine that runs mlpack.org might also work?
< rcurtin>
ah, yes
< rcurtin>
actually mlpack.org will be a better idea
< rcurtin>
because if we want mysql running on masterblaster I'll have to file a firewall change request and it could be weeks...
< zoq>
Okay, adding MySQL support was straightforward, we can now specify 'driver : "MySQL"' in the config file to use MySQL instead of SQLite. SQLite is still the default setting.
pantsforbirds has joined #mlpack
< rcurtin>
okay, I will get the right mysql databases set up on mlpack.org sometime soon, but I won't be able to do it today