[mlpack] NumFOCUS summit report and ideas

Thu Nov 7 13:37:16 EST 2019

Hello,

> mlpack video meetings:

What about we create a shared document (e.g. google docs) and use that to manage
the meetings, if someone has a particular question or topic they like to discuss
they can write it down before the meeting. This allows us to prepare for certain
topics. We can also keep track of what was discussed in the meetings.

> mlpack chat / IRC:
> 
> Moving from IRC may be a good idea - I personally can't use it, let's say
> because of "1990's environment restrictions".

I think the idea here is to extend the current situation and not to abandon IRC.
So the way I see it, correct me if I'm wrong, if everything works as described
above, anybody can pick their favorite application and everything works
interchangeably.

> giving talks on mlpack:
> 
> mlpack talks/workshops - this is a very good point. I've been doing some of this
> and it really pays off. The amount of user conversion you get by doing this is
> superior to opportunistic discoveries of the library. And I truly believe we
> need more users and more engagement. After all, that's why we build mlpack,
> right? I would add that we also need talks and workshops that are *not* super
> technical. For new users (consumers of the library I mean), metaprogramming is
> not really attractive. Instead, these users are looking forward a
> simple-useful-easy "Hello World" that delivers instantaneous value. Take a look
> at scikit-learn for instance, it's all about examples and ready to consume
> snippets.

Agreed, maybe we can put a list together with events (local/global) as a
starting point.

> C++ notebooks:

I have a setup running that I can make accessible, Sylvain used Binder, which
has some limitations like 1-2GB of memory, kernel shutdown after 10 minutes of
inactivity. If I'm thinking about running neural network examples, we might run
into the mentioned limitations.

Thanks,
Marcus

> On 7. Nov 2019, at 13:11, Germán Lancioni <gmansoft at hotmail.com> wrote:
> 
> Hi folks,
> 
> Thanks Ryan for sharing this. I agree with several ideas. A couple of notes:
> 
> 1) Moving from IRC may be a good idea - I personally can't use it, let's say because of "1990's environment restrictions".
> 2) mlpack talks/workshops - this is a very good point. I've been doing some of this and it really pays off. The amount of user conversion you get by doing this is superior to opportunistic discoveries of the library. And I truly believe we need more users and more engagement. After all, that's why we build mlpack, right? I would add that we also need talks and workshops that are *not* super technical. For new users (consumers of the library I mean), metaprogramming is not really attractive. Instead, these users are looking forward a simple-useful-easy "Hello World" that delivers instantaneous value. Take a look at scikit-learn for instance, it's all about examples and ready to consume snippets.
> 3) Website - This is our presentation card. That's why I hope the new website we are finishing will give us the status mlpack deserves. Again, better website and documentation brings more users. More users means also more contribution and more grow. That is the final satisfaction. Looking forward to working more on this.
> 
> Regards,
> German
> 
> From: mlpack <mlpack-bounces at lists.mlpack.org> on behalf of Ryan Curtin <ryan at ratml.org>
> Sent: Wednesday, November 6, 2019 07:44 PM
> To: mlpack at lists.mlpack.org <mlpack at lists.mlpack.org>
> Subject: [mlpack] NumFOCUS summit report and ideas
>  
> Hey everyone,
> 
> This past weekend Marcus and I attended the NumFOCUS summit
> (https://numfocus.org/summit2019 <https://numfocus.org/summit2019>) in New York City, since we are now a
> part of NumFOCUS. :)  I introduced mlpack to the rest of the NumFOCUS
> community with these slides:
> 
> https://www.ratml.org/misc/mlpack-numfocus.pdf <https://www.ratml.org/misc/mlpack-numfocus.pdf>
> 
> The summit was a great time, and I hope in future years that more of us
> will be able to attend.  I would say we are still figuring out the
> NumFOCUS thing at the moment.  They are too---NumFOCUS still has some
> other legwork to add until everything is set up and mlpack is shown on
> their website.
> 
> There were quite a lot of discussions over the weekend that I had, and I
> thought it would be useful to bring some of the ideas here for further
> discussion.  So, without further ado, I've split each thing into its own
> section.  If you're interested, please read on, and if you have any
> opinions or thoughts, please respond!
> 
> -----
> 
>   mlpack video meetings
> 
> I talked to a lot of people about how they used video chat and meetings.
> We've had meetings in the past and I think they've been successful, but
> one of the reasons they haven't happened as much is that I've prepared
> slides and other materials in the past, so it's high overhead, and I
> don't always have the time to do that regularly.
> 
> One idea I heard in discussions sounded pretty interesting to me: the
> AstroPy community has weekly calls, but they don't have a particular
> agenda and it's quite informal.  What I was told is that they all just
> jump on a call at a certain time each week, and whoever is there
> discusses PRs or directions or other ideas, or, alternately, people just
> write code while together in the same video chat room. :)
> 
> I think this is an interesting idea, and if others think so too, maybe
> we can try it!
> 
> -----
> 
>   mlpack chat / IRC
> 
> Both at the GSoC mentor summit and the NumFOCUS summit, I found that
> lots of communities use lots of different tools.  There's a huge
> proliferation of different chat services... we are maybe a bit
> antiquated with just IRC.  I've thought that for a while, but someone
> showed me Matrix (https://matrix.org <https://matrix.org/>) and I was quite impressed: Matrix
> has support for bridges to all kinds of different services.
> 
> So, as an experiment, I've connected Matrix to the mlpack IRC room.  If
> you join the Matrix room #mlpack:matrix.org, any messages you send there
> will get relayed to IRC.  There is also Gitter and Slack integration,
> which I think could be really useful---a lot of people in the past have
> asked about other chat services, but we haven't had anything to offer
> since in part my aim has been to keep things as low overhead as possible
> (and let's face it: with my plaintext emails and command-line mentality,
> I like to believe it's still the 90s...).
> 
> If you think this is interesting or a good idea, speak up, and maybe we
> can make the integrations better and then update the website to point
> people towards mlpack's chat services in a nicer way.
> 
> -----
> 
>   giving talks on mlpack
> 
> NumFOCUS runs the PyData conference, and I learned that PyData is not
> just about Python libraries.  So, it could be really nice to give talks
> at PyData about why C++ machine learning is possible, what things make
> it exciting (speed / template metaprogramming are my usual gotos), and
> how it can be done.  If you're interested in giving a talk on mlpack,
> NumFOCUS may be able to help, and I can definitely help with
> materials---I've prepared a lot over the years.
> 
> There are *lots* of opportunities to speak about what we're doing here,
> and I personally hadn't realized how many opportunities there are.
> 
> -----
> 
>   C++ notebooks
> 
> I spent some time talking with the CEO of QuantStack
> (https://quantstack.net <https://quantstack.net/>), who showed me a working C++ Jupyter notebook
> environment:
> 
> https://github.com/QuantStack/xeus-cling <https://github.com/QuantStack/xeus-cling>
> 
> It's built on top of the `cling` compiler, which is a modified version
> of the clang compiler from the ROOT project that can behave as a REPL.
> 
> Maybe people have known about this for a while, but this was the first I
> heard of it, and it's very exciting to me: in the past, a clear niche
> for mlpack was for production/deployment of models, and not so much
> interactive experimentation.  But xeus-cling seems like it will make it
> possible.
> 
> Personally, I'm going to be playing with xeus-cling and making some
> mlpack notebooks for it.  I'll share the efforts as they come along.
> QuantStack was really interested in being able to showcase a machine
> learning application inside of the xeus-cling notebooks, so there is a
> good amount of momentum and collaboration opportunity here.
> 
> (Just an aside: the xeus-cling notebooks currently suffer from problems
> whenever you try to redefine a variable---which happens a lot and is
> really inconvenient.  But the ROOT project released a new version of
> cling that has a workaround for this, so the xeus-cling maintainers just
> have to repackage that.  It may take a while, but that's the biggest
> showstopper that I saw for mlpack C++ notebooks, and if that gets fixed,
> I think we can have something really nice for people to interactively
> play with and use!)
> 
> -----
> 
>   logo design and website, etc.
> 
> When we came up with ensmallen, we did our best to find a logo, but it's
> a little bit generic.  I got connected to some graphic designers who
> might be able to help.  I have only very primitive logo design skills
> so maybe we can come up with something better here.  I'm not sure how it
> will work out, but I figured it was worth reporting. :)
> 
> -----
> 
>   mlpack/Shogun joint workshop
> 
> Each year, NumFOCUS helps Shogun put on a workshop, where their
> developers get together in one place.  I think this could be really
> great for mlpack, and maybe we could colocate a workshop like this and
> find some ways to collaborate between the two projects or something.
> 
> I'll probably send out a survey in a while to gauge interest in the
> idea, but of course, there would be a lot of planning to do and nothing
> is firm yet.  At this point we can consider it just a proposal, but I
> will say I think it would be super cool to get our development community
> in the same place!  For now we just run into each other occasionally at
> conferences and such.
> 
> -----
> 
> Okay, I know that I've written a lot!  It was a very informative weekend
> and I'm quite excited about the things that will come out of it.  If
> anything that you read above interests you, please feel free to respond!
> Like I said in the talk, mlpack is community-led, so, we all get to
> decide as part of a group what it is we think is cool and how we want to
> do it. :)
> 
> (And, if we disagree on direction, that's perfectly okay!  As part of
> our joining process, NumFOCUS asked us to come up with a governance
> policy to help guide how we make decisions, and we have a draft:
> https://github.com/mlpack/mlpack/pull/2068 <https://github.com/mlpack/mlpack/pull/2068> )
> 
> -- 
> Ryan Curtin    | "You got to stick with your principles."
> ryan at ratml.org |   - Harry Waters
> _______________________________________________
> mlpack mailing list
> mlpack at lists.mlpack.org
> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack <http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack>
> _______________________________________________
> mlpack mailing list
> mlpack at lists.mlpack.org <mailto:mlpack at lists.mlpack.org>
> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack <http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20191107/ec2c0acc/attachment-0001.html>