ChanServ changed the topic of #mlpack to: "mlpack: a fast, flexible machine learning library :: We don't always respond instantly, but we will respond; please be patient :: Logs at http://www.mlpack.org/irc/
KimSangYeon-DGU has joined #mlpack
< KimSangYeon-DGU>
rcurtin: I managed to wrap the lines to under 80 characters in the `.appveyor.yml`. :)
vivekp has joined #mlpack
< KimSangYeon-DGU>
zoq: The logs for 21th, 22th and 23th seem to be missing.
soonmok has joined #mlpack
< rcurtin>
KimSangYeon-DGU: I saw that, I'll review when I have a chance
< rcurtin>
not sure why the IRC logs keep failing to rebuild...
caiojcarvalho has joined #mlpack
< KimSangYeon-DGU>
Thanks. :)
soonmok has quit [Remote host closed the connection]
< zoq>
rcurtin: It looks like jekyll just ignores the page, what about we redirect on the server?
vivekp has quit [Ping timeout: 245 seconds]
vivekp has joined #mlpack
soonmok has joined #mlpack
< rcurtin>
the redirect works, if you go to ensmallen.org/demo.html
ayesdie has joined #mlpack
< zoq>
right, not sure why the navbar isn't showing the demo page in the navbar, my only guess is that it has something todo with the redirect plugin. The redirect works, so I guess a workaround would be to create a demo page without the redirect plugin and do the redirect on the server.
< ayesdie>
hey, I wanted a bit insight on why the Travis build for my PR was failing on OSX. I went through through the logs and the place where the build fails seems to build fine on other os.
< ayesdie>
also, I updated the `SparseSVMFunction::Evaluate()` and `SparseSVMFunction::Gradient()`. I anyone could review those changes, I can make some more tests on the SparseSVM.
< rcurtin>
zoq: we can do that if you like
< rcurtin>
ayesdie: I was hoping to look through your PR in a day or two, and I'll see if I have any insight on the failing travis build then
< zoq>
rcurtin: If that works, let's see what happens if I remove the redirect.
< rcurtin>
sure. let me see if I can set up the apache configuration
< rcurtin>
ok... redirect should be working
< rcurtin>
if you push to the ensmallen.org repo, it will auto-rebuild via a webhook for Jenkins
< rcurtin>
(and auto-deploy)
< zoq>
yeah, already did that; it works, thanks!
< rcurtin>
awesome
< rcurtin>
the CSS looks just a little different on firefox on vis.ensmallen.org, by the way
< rcurtin>
not a huge deal
< zoq>
I'll see if I can fix that as well.
< zoq>
ah, I see what you mean
ayesdie has quit [Ping timeout: 256 seconds]
KimSangYeon-DGU has joined #mlpack
_KimSangYeon-DGU has quit [Quit: Page closed]
cult- has joined #mlpack
< cult->
would it be easier to port armadillo to GPU with OpenACC instead of Cuda and OpenCL ?
< cult->
> referring to Bandicoot project
< rcurtin>
cult-: maybe, I am not familiar with OpenACC
< rcurtin>
it looks like it would be relatively easy, but I don't know how fast the resulting code would be compared to CUDA/OpenCL
soonmok has quit [Remote host closed the connection]
< cult->
"time to market" has a clear advantage here in my opinion
< cult->
as Bandicoot will hardly be in a production state anytime soon
< cult->
what other routines could be accelerated inside mlpack?
< rcurtin>
Bandicoot is my personal #1 priority for the year; I agree that time-to-market is a big deal here
< rcurtin>
it's sat for way too long
< rcurtin>
(but admittedly, it is a free time project...)
< rcurtin>
I'm not sure what could be accelerated in mlpack with OpenACC but honestly I would rather devote the time I do have to bandicoot
< rcurtin>
(not that I would oppose someone looking into OpenACC for mlpack. it could be really effective. maybe a good place to start might be, e.g., logistic regression?)
< cult->
\o/ bandicoot it is then
< cult->
i will look into the source perhaps i could also contribute to speed up the dev
< cult->
just realized there are at least 2 other open source projects under this name
< cult->
maybe it would make sense to change name of the project to.. garmadillo ? :)
< rcurtin>
yeah, I see it is also an open-source python toolbox for mobile phone metadata
< rcurtin>
I doubt conrad would be excited about a name change
< rcurtin>
but if so, he'd probably pick some other strange australian animal :)
< rcurtin>
basically, my current holdup is that I have to get some mlpack things off my plate first to make room for bandicoot. I'm hoping to have that done within a couple weeks
< rcurtin>
and then I was going to sync with Conrad, make a roadmap to a first release, and get to work
< rcurtin>
once we have that, maybe it will be clear how the work can be split up and done by many people to get it done faster :)
< cult->
awesome!
< rcurtin>
honestly the coding part is not that incredibly difficult---it's mostly just a wrapper, so the key is just getting the template metaprogramming framework in place to call the right CUDA/OpenCL functions
< rcurtin>
and that framework mostly already exists as a part of Armadillo
< rcurtin>
now I don't want to give the idea that it will only take a week of course. work always expands to fill all available time :)
< cult->
yes i just started to read it and i don't see any cuda or opencl stuff :o
< rcurtin>
it currently wraps, I think, CLMagma or CLBLAS
< rcurtin>
I would probably have to write the CUDA wrapper support, but I definitely think that's necessary due to nvidia's dominance in the space
< cult->
do we need cuda when opencl is basically compatible with nvidia card?
< rcurtin>
I would say so---I would expect CUBLAS to be faster than any OpenCL-equivalent implementation
< rcurtin>
I picked up an RTX 2080ti a few weeks ago so I can do some testing
< cult->
why moved to gitlab by the way?
< rcurtin>
so I will know for sure once I do that
< rcurtin>
Conrad didn't like the Microsoft acquisition of Github so he moved it
< rcurtin>
Gitlab has some nice features, I have no issue with it. I think it would be too much work to move mlpack at this time though
< cult->
github is still #1 for open source but fine
< rcurtin>
agreed
< rcurtin>
but I do think Gitlab is #2 now :)
< rcurtin>
(definitely better than what sourceforge has become)