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/
< rcurtin>
hello there everyone, I think I will wait another minute or two for a few people that I haven't seen yet
< zoq>
yeah, good idea :)
< Kirill>
hey guys
govg has joined #mlpack
< chenzhe>
hi!
< zoq>
Hello there!
< rcurtin>
hey there Kirill, glad you could make it :)
< rcurtin>
I thought maybe I had sent the invitation to the wrong email since I didn't see you this morning, but I guess you got it in the end :)
< Kirill>
yeah, I did
< rcurtin>
ok, I guess I will go ahead and start, people can read the logs if they need to catch up :)
< rcurtin>
so, hello everyone, thanks for coming to the meeting! this is the closest we can get to an in-person meeting since we are all so far apart
< rcurtin>
congratulations to the students on your acceptance!
< rcurtin>
as you probably know, this is the second instance of this meeting, so basically I'll be talking about the same things
< rcurtin>
but for some reason the log is not fully working, so I can't seem to load the messages from earlier today... I will have to look into that...
< rcurtin>
anyway, I think you all know me, I'm Ryan Curtin, the mlpack GSoC organization administrator
< chenzhe>
It seems to be quite a long log to read ^_^
< rcurtin>
yeah, many messages :)
< rcurtin>
this is the biggest GSoC mlpack has had; this year we have 10 students. In previous years we've had 6, 5, and 3... so this is a much bigger logistical challenge!
< rcurtin>
I believe I have sent everyone an email with useful links, documentation, and other information about the project; if you didn't get that email, let me know and I'll send it to you (and make sure I send everything to your correct email in the future)
< rcurtin>
as far as schedule goes...
< rcurtin>
right now is the "Community Bonding" period, which goes from May 4 to May 30
< rcurtin>
ideally, during this time, you can get to know your mentor a bit, get to know the community a bit, maybe have some fun, maybe learn a bit more about mlpack
< rcurtin>
some nice ways to do this are on the mailing list, or direct emails to your mentor, or here on IRC---don't feel restricted to talk only about mlpack; we can have some fun in the channel also :)
< rcurtin>
once the community bonding period is over, the actual coding goes from may 30 to august 21
< rcurtin>
during that time, there will be two midterm evaluation periods; one at the end of June, and one at the end of July
< rcurtin>
then, at the end of the summer there's a final evluation
< rcurtin>
after that, there's no schedule imposed by Google, but we hope that you will stick around and continue to participate in the project :)
< rcurtin>
if there are any stipend issues or administrative issues, probably Google will be the most helpful there but you should feel free to ask me and I can try and help or escalate to Google as needed
< rcurtin>
so, before I moved on, any questions about schedule or anything? next we can talk about student and mentor expectations
< rcurtin>
ok, I will take that as a no :)
< Kirill>
yeah, we got it
< rcurtin>
I don't think that any of the expectations here are too difficult, but I think it's important to discuss these before the summer starts to ensure we are all on the same page
< rcurtin>
students are expected to work the equivalent of a full-time job or internship; so, a full work week
< rcurtin>
it's okay if some weeks you work more and some weeks you work less, but in the end it should even out
< rcurtin>
it's also okay if you have to travel or will be unavailable, but please make sure your mentor knows that you'll be gone
< rcurtin>
disappearing students are a big problem in GSoC, so if we don't hear from you for a little while we may start to get scared that you have disappeared :)
< rcurtin>
regular contact and communication with mentors is very important and expected; preferably, this would be via the public #mlpack IRC channel, but if you or your mentor prefer alternate means of contact that is okay too
< rcurtin>
the reason we suggest using the public channel or public mailing list is so that more people than just your mentor can answer any questions; sometimes this can be helpful
< rcurtin>
if you are having trouble with some part of your project, definitely do not be afraid to ask---that is what the mentor is there for
< rcurtin>
students are also expected to provide some kind of weekly status update to the community
< rcurtin>
this could be an email post, like these examples:
< rcurtin>
whichever way you'd like to do it is up to you, but weekly updates are important because people in the community may be interested in following what you are up to in your project
< rcurtin>
and I'll make sure you have the right permissions to post there after the meeting
govg has quit [Ping timeout: 240 seconds]
< rcurtin>
not every project goes according to plan; sometimes, the project may fall way behind the timeline, or it may proceed way ahead of the timeline
< rcurtin>
this is not necessarily a problem, do not worry!
< rcurtin>
if this does happen, the student and the mentor should discuss to see what is realistically accomplishable in the rest of the summer and adjust the goals accordingly
< rcurtin>
many GSoC projects don't get completed in the original way they were defined, so don't worry if things change during the summer---estimating how much work some software can take is a very hard task, and almost nobody gets it right
< rcurtin>
we really hope we won't fail any students this year, and we certainly don't expect to---every accepted student is a good student, so the prior probability of failure is very low in our opinions
< rcurtin>
but, we will fail a student who disappears, doesn't appear to be working the expected amount, or who is otherwise seriously underperforming
< rcurtin>
if such a situation does occur, the student will be made fully aware with warnings, so any failure will not be unexpected
< rcurtin>
like I said, I really don't expect this to be an issue, but it's important to talk about it beforehand in case any issues actually do arise, so that we are on the same page
< rcurtin>
any questions about student expectations? if not, I'll go ahead and move onto mentor expectations
< zoq>
nothing from my end :)
< rcurtin>
this meeting is going faster than the last one, not very many questions :)
< rcurtin>
no complaints from my end about that :)
< Kirill>
:)
< rcurtin>
next I'll talk about mentor expectations, which are also important to set out beforehand
< rcurtin>
mentors should work with their students to determine times that they are both available to work together
< rcurtin>
they should also be willing to debug code problems and help the student understand any theory as needed
< rcurtin>
similar to how students are expected to be available and in contact with their mentors, mentors are also expected to be responsive and in contact with their students, providing reasonably quick replies
< rcurtin>
the mentor shouldn't do the majority of the work on the project, of course, but they should be there to help as needed
stephentu has joined #mlpack
< rcurtin>
like earlier, the best form of communication for a student-mentor pair is going to depend on the preferences of the student and mentor, but ideally this may be best to do in public, to allow others to contribute/observe/ask questions if they might like
< rcurtin>
hey Stephen :) if you want to catch up, some logs are here: http://mlpack.org/irc/
< rcurtin>
(or maybe you already read the logs from this morning, in which case, you probably know what I'm about to say! :))
< rcurtin>
the two midterm evaluations and final evaluation need to be done by the mentor during the periods described by Google (end of June, end of July, and end of August)
< rcurtin>
if there is some problem, I think I can enter the evaluations as an organization administrator
< rcurtin>
but as always, if there's any problem you can just ask me (or zoq or whoever) and we can all try and get it figured out
< rcurtin>
any questions about mentor expectations?
< rcurtin>
also, can everyone send me the github account they'd like to use this summer so I can add them to the right Github team for GSoC students?
< Kirill>
ok
< stephentu>
i'm glad i joined around the time you were talking about mentor expectations lol
< rcurtin>
:)
< stephentu>
sorry i had a meeting run late
< Kirill>
here or throug mail?
< rcurtin>
no problem, don't worry about it
< rcurtin>
here is fine, you can just give me the ID
< Kirill>
micyril
< rcurtin>
Kirill, I think I already have yours, I'm assuming you'll use the same one you have in the past
< rcurtin>
yeah, right, let me add that now
< rcurtin>
and Stephen you're already a member of the organization :)
< Kirill>
yeah, I sent it just to make sure
< rcurtin>
chenzhe: what Github account id should I add for you?
< rcurtin>
I guess I need Kartik's also, but I think he is not here
< rcurtin>
anyway, those can be added later, no need to do it now :)
< rcurtin>
next I'll add some short history, I dunno if it will be interesting for anyone, but I think it is interesting :)
< rcurtin>
mlpack was first developed in 2007 in a lab at Georgia Tech (so, over ten years now!)
chenzhe1 has joined #mlpack
< rcurtin>
the lab had maybe ~10 people that contributed to the library early on, and I joined the effort around late 2009
< rcurtin>
my job, when I joined the lab, (in addition to research...) was to prepare the library to actually release as open source
< rcurtin>
but this took two full years of refactoring and a team of people, so mlpack 1.0.0 wasn't released until december 2011 at a NIPS workshop
chenzhe has quit [Ping timeout: 246 seconds]
chenzhe1 is now known as chenzhe
< rcurtin>
at that point, the lab kind of died when the advisor (Alex Gray) left Georgia Tech to start a company called Skytree
< rcurtin>
and the few remaining people (which I think at some point was just me) got involved with Google Summer of Code, and the community has grown a ton since then
< rcurtin>
this is our fourth Summer of Code, and like I said earlier by far the biggest
< rcurtin>
now the library has somewhere over 80 contributors, from all different continents except Antarctica
< rcurtin>
and one of the contributors is actually a deep learning system: https://github.com/C0deAi
< rcurtin>
there's also a pull request open from North Korea, so I think we are the only ML library with code that's come from there :)
< Kirill>
:D
< rcurtin>
it is very exciting that when I travel to conferences now, there is some good name recognition of mlpack---people know what it is, unlike in 2012
< rcurtin>
I'm really hoping that many of the projects this summer will get a lot of interest from the larger machine learning community
< rcurtin>
many of the projects are focused on the neural network code, which I am hoping we will be able to release as stable soon, and that will probably get a lot of interest
< stephentu>
can i ask a general mlpack question
< rcurtin>
in addition, I am currently working on automatic Python bindings for the command-line programs, and this should help bring more people to use mlpack (and maybe other languages too)
< rcurtin>
of course, go ahead
< stephentu>
how do you see mlpack w/ respect to all the other ML frameworks out there like sklearn, and all the DL frameworks like pytorch adn tensorflow
< stephentu>
esp the ones w/ company backed support
< rcurtin>
I think the focus is pretty different, or, at least it traditionally has been different
< rcurtin>
technically we have a little bit of support via Symantec, but that's not the same thing as, e.g., Spark and Databricks :)
< rcurtin>
traditionally mlpack focused on very fast implementations instead of ease of use
< rcurtin>
since it's in C++, it's already much higher on the learning curve than most people want to climb
< rcurtin>
mlpack also has typically focused on less "standard" algorithms, and has implementations of a lot of stuff you won't find elsewhere
< rcurtin>
(though we have definitely added more 'standard' techniques over the years)
< stephentu>
ya i think thats how i stubmled upon mlpack in teh 1st place
< stephentu>
i was looking for some SDP implementation
< rcurtin>
yeah, that's a decent example---mlpack has one of the better optimizer frameworks out there (in my opinion)
< stephentu>
cool
< rcurtin>
I think, moving forward, that the best path might be to focus on speed, then ease of use
< Kirill>
does mlpack supports running in parallel on multi-core processors?
< rcurtin>
Kirill: sort of; there is some OpenMP support for some algorithms
< rcurtin>
and you could also use OpenBLAS inside of Armadillo to get parallel linear algebra
< rcurtin>
Shikhar's project this year will focus on some parallelization too
< Kirill>
ok
< rcurtin>
that's not to say it's perfect; there is a lot of room for improvement :)
< rcurtin>
but, I think maybe "there is a lot of room for improvement" applies to just about any code anywhere :)
< rcurtin>
stephentu: back to your original question, I'm not fully sure how to see mlpack's DL code vs. pytorch or tensorflow yet
< rcurtin>
I have to come up with something, because deep learning is really hot inside of Symantec and my goal is to get people to use mlpack inside of Symantec :)
< rcurtin>
so I will have to figure out some persuasive arguments for why one would use mlpack vs. TF or whatever else... and I think maybe benchmarks will be a good part of that argument
< stephentu>
good luck, its a tough battle to fight
< rcurtin>
agreed, it absolutely is
< rcurtin>
but in the mean time, I am having fun working on mlpack and I think there are places where we definitely provide some nice support that other libraries don't :)
< rcurtin>
so I don't have too much insight on "where things will be" in a year or so, we will have to see, but I do know that the improvements from this summer's projects will be exciting and my opinion is they will capture a good amount of interest :)
< rcurtin>
I guess, I'm out of things to say in the meeting, but I think maybe it is a good idea to have some introductions so we can get to know each other a little bit
< rcurtin>
I don't have any formal structure, so maybe it will be chaotic, but...
< rcurtin>
I'm Ryan, I live in Atlanta, I did my B.S., M.S., and Ph.D. at Georgia Tech, and now I work for Symantec and still managed to stay in Atlanta
< chenzhe>
it seems that I just loss connection when you ask me about github account, did you get it?
< rcurtin>
chenzhe: no, I didn't see the message, sorry about that
< Kirill>
So, maybe it's a good time to introduce myself
< chenzhe>
Looks similar, this city is actually famous for all kinds of noodles
< chenzhe>
sure
< Kirill>
my name is Kirill as you can guess :)
< Kirill>
I'm a PhD student at Ural Federal Univercity (Ekaterinburg, Russia) working on Computational Humor
< Kirill>
This summer I'm going to work on cross-validation and hyper-parameter tuning infrastructure
< stephentu>
Kirill: will you be experimenting w/ any of these bayesian methods for hyperparam selection?
< Kirill>
In my free time I like to walk and cycle with friends
< Kirill>
stephentu: it will be beyond the scope of this summer, but it can be extanded in this in the future
< Kirill>
I hope to make it flexible enough to make it possible
< stephentu>
cool, ya its definitely a lot of work to implement that stuff
< rcurtin>
yeah, I am very excited about this project, but you probably already know that from the emails :)
< stephentu>
so rcurtin can you explain a bit more about what community bonding entails
< stephentu>
maybe i missed this
< stephentu>
we're not techically supposed to get started yet
< stephentu>
from what i can tell
< rcurtin>
yeah, the students don't need to write any code yet (unless they want to)
< rcurtin>
the idea is for students to integrate into the community and get to know the other students and mentors
< rcurtin>
so that the summer itself is more fun and feels less like a heartless consulting job :)
< stephentu>
haha
< rcurtin>
in a real job I guess this would be the equivalent of water cooler discussions
< rcurtin>
but we don't have an international water cooler, just #mlpack and the mailing list :(
< rcurtin>
(and whatever other communication methods)
< rcurtin>
there are no strict and hard requirements for the community bonding period though, just a general idea :)
< rcurtin>
ok, I guess there is nothing else for now, maybe we will hear from Kartik and Sumedh in the future :)
< rcurtin>
feel free to idle in the channel and chat!
< chenzhe>
Sure
< rcurtin>
thank you everyone for attending the meeting
< rcurtin>
I'll be around on and off for the next couple hours before I go to bed
< rcurtin>
I am looking forward to the summer :)
Kirill has quit [Ping timeout: 260 seconds]
< stephentu>
great thanks
< stephentu>
i'll try to come on irc more
stephentu has quit [Quit: Lost terminal]
sumedhghaisas has joined #mlpack
mikeling has joined #mlpack
chenzhe has quit [Ping timeout: 260 seconds]
govg has joined #mlpack
govg has quit [Ping timeout: 240 seconds]
vivekp has quit [Ping timeout: 272 seconds]
vpal has joined #mlpack
vpal is now known as vivekp
mentekid has joined #mlpack
sumedhghaisas has quit [Ping timeout: 240 seconds]
sumedhghaisas has joined #mlpack
govg has joined #mlpack
sumedhghaisas has quit [Ping timeout: 240 seconds]
< rcurtin>
ok, I set up the benchmarks repository to post to the mlpack-git mailing list when commits are pushed
< rcurtin>
I'll also get PRs and issues set up for the other repositories to send emails to the mlpack-git mailing list
< zoq>
good idea and there is the first one
< rcurtin>
ok, that should be set up correctly now
< rcurtin>
now since I am on the mlpack-git mailing list I have to now set my personal github account to ignore all activity on the blog and benchmarks repositories
< zoq>
yeah, a new filter for me as well
sumedhghaisas has joined #mlpack
chenzhe has joined #mlpack
mikeling has quit [Quit: Connection closed for inactivity]
chenzhe has quit [Ping timeout: 260 seconds]
chenzhe has joined #mlpack
sumedhghaisas has quit [Ping timeout: 240 seconds]