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/
utopiar has joined #mlpack
< utopiar>
hello everybody, i'm a newbie. I want to ask a question: How can i compile the module "ann" in the methods?
< utopiar>
there is no a cmake file, i try touch one, but i think it required c++14 or higher version?
utopiar has quit [Ping timeout: 246 seconds]
< naywhayare>
oops, got back just a second too late to answer the question... :(
plutocracy has joined #mlpack
< plutocracy>
hello
< naywhayare>
plutocracy: hello there
< plutocracy>
it's conrad here.. how are you doing?
< naywhayare>
good, I just finished turning in my paperwork, so I'm officially finished :)
< naywhayare>
I suspect you are here about the Armadillo EPEL package... :)
< plutocracy>
nice. congrats!
< naywhayare>
thanks... I brought the correct type of coffee and donuts (no alcohol allowed, unfortunately, due to campus policies), and as a result none of the committee had any issues with the work
< plutocracy>
nah, mainly just to have a general chat.
< naywhayare>
ah, about plutocracy and other forms of government, perhaps? :)
< plutocracy>
awesome. so the thesis is all approved, or does it have to through external examiners?
< plutocracy>
plutoracracy as according to the infamous Citi bank memo :-)
< naywhayare>
ah, I see
< naywhayare>
the thesis is approved by the committee, and the only people who need to look over it now are the format checkers
< naywhayare>
who will inevitably have some horribly minor issue with the numbering of footnotes or the particular spacing of two words or something else trivial
< plutocracy>
nitpicky. they hammer us over here with external examiners, but they leave the format up to us. I wonder if that's on oz thing, or inherited from the UK
< naywhayare>
my impression here is that nobody actually has the time to read the thesis, so comments are generally quite sparse
< naywhayare>
the only comment I received before the defense was "it was long"...
< plutocracy>
that's actually a good comment !
< plutocracy>
they probably secretly weighed the thesis and realised they had to carry it around
< plutocracy>
so what's next for you?
< naywhayare>
I'm joining Symantec as a research scientist, which should be interesting
< naywhayare>
they are starting a new machine learning department
< naywhayare>
what they told me is that half my time can be devoted to mlpack/continuing my research, and the other half to applying mlpack/research to Symantec internal problems
< naywhayare>
so, that's kind of a deal I didn't want to pass up, and it also lets me stay in Atlanta, which is nice
< naywhayare>
fortunately I will also have no affiliation with their antivirus software... :)
< plutocracy>
brilliant!
< plutocracy>
money should be decent too?
< naywhayare>
yeah, it's better than a postdoc, that's for sure :)
< naywhayare>
I saw the horrendous hours and stress that my colleagues who went that direction work, and I lost interest...
< naywhayare>
I'm told I'll be sent to ICML and NIPS each year (or other conferences), so it should be quite enjoyable to be able to go and occasionally publish without the kind of pressure postdocs and assistant professors are under
< plutocracy>
yeah, postdoc is pretty much PhD part 2. you quickly end up going in the academic direction by maximising the amount of published papers (often via the salami approach)
< plutocracy>
it's good to be on the cross roads btwn industrial and academic research. I find it far more satisfying, as you can still work on interesting problems that get applied to real life stuff
< naywhayare>
yeah, I am quite happy with what I think the position will be (so hopefully it will turn out to be what I think it is)
< plutocracy>
things here are a bit of a mess though. we're going through an alleged merger with csiro (a massive national science org). supposed to happen any minute now, but there's always one more signature to be done by someone
< plutocracy>
if things don't work out, we're going to shut down mid next year
< naywhayare>
!! sorry to hear that
< naywhayare>
I would like to hope that NICTA is a large enough organization with enough momentum that a shutdown won't be allowed to happen
< naywhayare>
I'm reading about the merger now... this sounds like bad news :(
< plutocracy>
there's a new ceo for all of csiro, who has many years experience in the silicon valley. everyone is waiting for the penny to drop and have everything re-arranged. it might be good though to have a new focus
< plutocracy>
i have another project in the pipeline which is going to complicate things a bit
< plutocracy>
involves lots of screaming, and changing nappies. due in about a month
< naywhayare>
ah, congratulations!
< plutocracy>
thanks :-)
utopiar has joined #mlpack
< plutocracy>
so I might be "offline" so to speak from about mid-september, for a month or two.
< naywhayare>
yep, I can imagine
< naywhayare>
I'm hoping to have time to finally finish the SuperLU project, but with the thesis defense, I just completely ran out of time
< naywhayare>
the GSoC student has (mostly) finished a reimplementation of ARPACK, too, so once I benchmark and test that, I should send you an email with the results and see if you're interested in merging in the resulting project
< plutocracy>
yeah, no worries. I've come to realise that there will always be something to do with Armadillo -- it's one of those infinite projects
< naywhayare>
yeah, I feel the same way working on mlpack (and my research too)... I have orders of magnitude more ideas than time
< plutocracy>
a new ARPACK would be good. does it implement all the current options for ARPACK?
< naywhayare>
I believe that it does, yes; or, at the very least, it implements all that's needed to replace ARPACK in eigs_gen() and eigs_sym()
< utopiar>
hello everybody, forgive me, i'm a newbie. I want to ask a question: How can i compile the module "ann" in the methods?
< naywhayare>
hi utopiar, zoq just responded to your bug on github... he mentioned he can send you the CMakeLists.txt files, and that he's currently refactoring the code
< naywhayare>
I don't know anything more than that though :)
< utopiar>
OK, I'm so excited with that. I'm so love your work!
< naywhayare>
thanks! I hope that it is useful for you
< naywhayare>
please feel free to file bugs or ask questions in IRC if you have any issues in the future
< plutocracy>
I guess we can give the user a switch to choose between "standard" arpack or the replacement arpack, with the caveat that the replacement is experimental and/or doesn't support all the options. however, it would still be good to support the options ("lm", "sm", ...) so it can eventually totally supplant the standard arpack.
< naywhayare>
yep, I know it does support all the different options for which eigenvectors are recovered
< naywhayare>
it would be easy to build an interface around it so that its API is identical to ARPACK's (or vice versa)
< plutocracy>
nice. octave and scilab might be interested in this too, as to my understanding they're basically keeping arpack on life support (in the arpack-ng guise)
< plutocracy>
so are you gearing up for an update to mlpack?
< naywhayare>
I'll pass on that info to the student, and see if he is interested in pushing his work to those projects; it's apparently already in use by the rARPACK project
< naywhayare>
I'm hoping to release mlpack 1.1.0 in the next handful of weeks; I need to do some documentation work and API refactoring
< naywhayare>
some of the abstractions we originally used have outlived their usefulness and make things like loading/saving models very hard (or downright impossible)
< plutocracy>
how about calling it 2.0? It seems that it has lots of new features and changes. a shiny new version number would raise more interest too
< naywhayare>
it's a possibility; I haven't nailed it down yet
< naywhayare>
at this point that might be reasonable though, given that the scope of these API changes is broader than I anticipated
< plutocracy>
yeah, 2.0 would also be a good indicator that API changes are expected, so people don't complain when their code breaks
< naywhayare>
one nice thing is that it may end up being header-only, preventing bizarre issues where ARMA_64BIT_WORD was enabled during mlpack compilation but is not during the compilation of something that links against mlpack
< plutocracy>
going through mlpack commits, i've noticed that its codebase is deliberately not using new armadillo features. I think all relevant linux distros already have fairly recent versions.
< plutocracy>
i'll check now
< naywhayare>
there are definitely some cases where we've maintained reverse compatibility
< plutocracy>
mac os x macports: 5.400.2
< naywhayare>
but that has more to do with the fact that the mlpack build servers test on as far back as 3.6 I think?
< plutocracy>
mac os x homebrew: 5.200
< naywhayare>
I'll probably update all of those soon, but, unless I'm overlooking something (which I probably am), I don't think we're failing to get the most out of Armadillo
< naywhayare>
actually if you have any examples, I'd be more than happy to look into them
< plutocracy>
can't remember from the top of my head. it certainly raised an eyebrow, so it must have been some feature which was in 4.x for ages
< naywhayare>
ah, okay; if you do remember, let me know. it's completely possible that I have some habits of writing very archaic Armadillo code :)
< plutocracy>
the lowest version among the support crop of linux distros is 4.200. mainly ubuntu 14.04 and debian 7
< naywhayare>
I should definitely update the minimum required Armadillo version... it cuts down on the amount of work the build slaves have to do, too
< naywhayare>
like I said, far more ideas than time :)
< plutocracy>
yeah, it never ends
< plutocracy>
had a look at the commit. norm() is one of them, but I don't think it was that one. It was probably the specification of direction in sort(). rather than using 0 or 1, you can now use "ascend" or "descend", which is more intuitive
< naywhayare>
ah, that's probably related to my own archaic internally memorized Armadillo API :)
< plutocracy>
the old API still works, so you can certainly keep on using it. I'm not planning on removing it, but I recommend for new code to use the easier to digest api :)
< naywhayare>
yes, definitely
< naywhayare>
I've finally transitioned my own usage from trans() to .t(), but there's lots of legacy code
< plutocracy>
btw, that other thing with the certain south-western mob: no news on that front
< naywhayare>
what, Cliven Bundy? I haven't heard anything about him in a while
< naywhayare>
or maybe I missed the reference :)
< plutocracy>
not that one :-) I'll send you an update via email
< naywhayare>
ah, okay
< plutocracy>
ok, I think the new episode of Mr Robot has arrived :-)
< naywhayare>
hmm, I'd never heard of that show. I think now that I'm done being a graduate student, I'm allowed to climb out from under my rock and participate in culture again
< plutocracy>
it's pretty decent, bit of a head trip too. you're looking at the world through his eyes, and he chooses to be a deceptive narrator.
< naywhayare>
ah, fascinating
< naywhayare>
I will add it to the list :)
< plutocracy>
ok. I'm going to head off. good to chat :-)
< naywhayare>
right, talk to you later
plutocracy has left #mlpack []
< zoq>
utopiar: I just responded to your issue on github, let me know if you have any further questions or issues.