Re: Additional mentor to MXNet - Jim Jagielski

[Steffen Rochel]

Welcome Jim, appreciating  your support.
Steffen

On Mon, Jun 18, 2018 at 3:14 PM Naveen Swamy  wrote:

> Hi All,
>
> I am excited to announce that we have an additional mentor for MXNet: Jim
> Jagielski, the cofounder of Apache, he brings vast experience in building
> and growing successful communities around projects. I am sure we will be
> able to tap into his experience to bring alignment and build a strong
> community around MXNet.
>
> He has a page on wikipedia, check it out :)
> https://en.wikipedia.org/wiki/Jim_Jagielski.
>
> Thanks, Naveen
>

Re: Additional mentor to MXNet - Jim Jagielski

[Hagay Lupesko]

That is great news!
Welcome on board Jim!


On Mon, Jun 18, 2018, 15:14 Naveen Swamy  wrote:

> Hi All,
>
> I am excited to announce that we have an additional mentor for MXNet: Jim
> Jagielski, the cofounder of Apache, he brings vast experience in building
> and growing successful communities around projects. I am sure we will be
> able to tap into his experience to bring alignment and build a strong
> community around MXNet.
>
> He has a page on wikipedia, check it out :)
> https://en.wikipedia.org/wiki/Jim_Jagielski.
>
> Thanks, Naveen
>

[VOTE] Release MXNet version 1.2.1.RC0 (Patch Release)

[Anirudh]

Hi,

This is the vote to release Apache MXNet (incubating) version 1.2.1. Voting
will start now and close Thursday June 21st 7:00 PM PDT.

Link to release candidate 1.2.1.rc0:

https://github.com/apache/incubator-mxnet/releases/tag/1.2.1.rc0

View this page for installation instructions:

https://mxnet.incubator.apache.org/install/index.html

(Note: The README.md points to the 1.2.1 tag and does not work at the
moment).

Please remember to test first before voting accordingly.

+1 = approve
+0 = no opinion
-1 = disapprove (provide reason)

Anirudh

Re: Nightly tests README accurate?

[Marco de Abreu]

Thanks for your input. Do these tests really require a p3 or would g3 be
sufficient? We got 2 GPUs on the g3 instances and they are a lot

-Marco

Indhu  schrieb am Mo., 18. Juni 2018, 15:02:

> I was working on testing the notebooks from https://gluon.mxnet.io/ as
> part
> of the nightly tests to make sure we notice any API change that could break
> code in those notebooks.
>
> The README in the nightly folder said the tests are being run on machines
> with multiple GPUs. I was wondering if I have to run multiple notebooks in
> parallel to efficiently utilize all GPUs. After reading your email, I'm
> thinking I'll run most of the notebooks on P3.2xl and only the ones that
> require multiple GPUs on P3.8xl. Thanks for the reply.
>
>
> On Mon, Jun 18, 2018 at 2:24 PM Marco de Abreu
>  wrote:
>
> > In future (next few days) the nightlies will be running on our regular CI
> > at http://jenkins.mxnet-ci.amazon-ml.com/
> >
> > On there, we support C5.18xlarge (Ubuntu, CentOS within Docker and
> > Windows), G3.8xlarge (Ubuntu, CentOS within Docker and Windows),
> P3.2xlarge
> > (Ubuntu, CentOS within docker) and (upcoming) P3.8xlarge (Ubuntu, CentOS
> > within docker).
> >
> > What exact requirements do you have, Indhu? I'm happy to assist!
> >
> > -Marco
> >
> > On Mon, Jun 18, 2018 at 1:47 PM Indhu  wrote:
> >
> > > Thanks Meghna. Do you know what hardware the nightly tests are
> currently
> > > running on? I'm adding some nightly tests. It will be useful to know.
> > >
> > >
> > > On Mon, Jun 18, 2018 at 9:11 AM Meghna Baijal <
> > meghnabaijal2...@gmail.com>
> > > wrote:
> > >
> > > > Hi,
> > > > This readme is incorrect and I am fixing this as a part of the
> nightly
> > > > tests PR soon.
> > > >
> > > > Thanks,
> > > > Meghna
> > > >
> > > > On Sun, Jun 17, 2018 at 7:05 AM kellen sunderland <
> > > > kellen.sunderl...@gmail.com> wrote:
> > > >
> > > > > I'm not a commiter, but I would say this is clearly not correct.  I
> > > would
> > > > > propose removing the doc until someone has time to verify that the
> > > steps
> > > > > work correctly, and that the descriptions are accurate.
> > > > >
> > > > > On Thu, Jun 14, 2018 at 10:04 PM Indhu 
> > > wrote:
> > > > >
> > > > > > Is the README
> > > > > > <
> > https://github.com/apache/incubator-mxnet/tree/master/tests/nightly
> > > >
> > > > > for
> > > > > > the nightly tests accurate? For example,
> > > > > >
> > > > > > 1. Are tests being run on machines with Intel i7-4790 and 4
> Nvidia
> > > GTX
> > > > > 970
> > > > > > Tis?
> > > > > > 2. Is http://ci.dmlc.ml/ the right place to look for build
> status?
> > > > > > 3. Is the instruction to run on Jenkins correct?
> > > > > >
> > > > > > If not, what all needs to be changed in that page?
> > > > > >
> > > > >
> > > >
> > >
> >
>

Re: Update on 1.2.1 release

[Anirudh]

Hi all,

Since there are some issues with publishing scala package to maven, the
1.2.1 vote doesn't have to be blocked for the Scala package release.
I will create the vote for 1.2.1.rc0 soon.

Anirudh

On Mon, Jun 18, 2018 at 1:27 PM, Anirudh  wrote:

> Hi Sina,
>
> RC is available here: https://github.com/apache/incubator-mxnet/
> releases/tag/1.2.1.rc0 . We are waiting to publish scala package to maven
> before we start the vote. Naveen, Qing and Andrew are working on this.
>
> Anirudh
>
> On Mon, Jun 18, 2018 at 1:18 PM, Afrooze, Sina <
> safro...@amazon.com.invalid> wrote:
>
>> Just wondering if there is any estimate on when a 1.2.1 RC may be
>> created? We are using 1.3.0 beta builds because those are the only
>> pip-install-able builds available, but they introduce some instability in
>> our workflow that we'd like to avoid by switching to a 1.2.1 build. - Sina
>>
>> On 6/15/18, 12:55 PM, "Marco de Abreu" 
>> 
>> wrote:
>>
>> Hello,
>>
>> we would need this PR to be merged as well:
>> https://github.com/apache/incubator-mxnet/pull/11309
>>
>> With the next deployment of new Windows slaves (which should happen
>> asap),
>> the 1.2 branch and all past branches are going to break since the test
>> execution is not versioned. Please merge this PR to ensure we're
>> still able
>> to test the 1.2 branch on our CI.
>>
>> -Marco
>>
>> On Thu, Jun 14, 2018 at 6:20 PM Anirudh 
>> wrote:
>>
>> > Hi all,
>> >
>> > We have one last PR before code freeze:
>> > https://github.com/apache/incubator-mxnet/pull/11298
>> >
>> > Anirudh
>> >
>> > On Thu, Jun 14, 2018 at 11:46 AM, Anirudh 
>> wrote:
>> >
>> > > Waiting on CI for the PRs: #11236, #11210, #11267
>> > >
>> > > Other PRs have been merged.
>> > >
>> > > On Wed, Jun 13, 2018 at 10:50 PM, Anirudh 
>> wrote:
>> > >
>> > >> Thanks Tao! Yes, this shouldn't be a blocker for 1.2.1.
>> > >>
>> > >> On Wed, Jun 13, 2018 at 10:46 PM, Lv, Tao A 
>> wrote:
>> > >>
>> > >>>
>> > >>> Yes, #10311 is only in master branch, so I guess it won't
>> impact 1.2.0
>> > >>> branch and block the release of 1.2.1, right?
>> > >>>
>> > >>> A PR (#11273) is submitted to disable the test temporally and
>> hopefully
>> > >>> it will be fixed soon.
>> > >>>
>> > >>> -tao
>> > >>>
>> > >>> -Original Message-
>> > >>> From: Marco de Abreu [mailto:marco.g.abreu@googlema
>> il.com.INVALID]
>> > >>> Sent: Thursday, June 14, 2018 1:21 PM
>> > >>> To: dev@mxnet.incubator.apache.org
>> > >>> Subject: Re: Update on 1.2.1 release
>> > >>>
>> > >>> On windows tests a segfault is indicated by "If -764728474728".
>> I have
>> > >>> also seen it happen on Ubuntu, there are probably some links in
>> the
>> > issue
>> > >>> (on my phone right now).
>> > >>>
>> > >>> Anirudh  schrieb am Mi., 13. Juni 2018,
>> 22:16:
>> > >>>
>> > >>> > By segfaulting test do you mean : test_gru_bidirectional. I
>> don't see
>> > >>> > the segfault in the logs. Can you point me to the test.
>> > >>> > Also, this seems to be specific to the master and not in 1.2:
>> > >>> > https://github.com/apache/incubator-mxnet/pull/10311
>> > >>> >
>> > >>> > Anirudh
>> > >>> >
>> > >>> > On Wed, Jun 13, 2018 at 10:00 PM, Marco de Abreu <
>> > >>> > marco.g.ab...@googlemail.com.invalid> wrote:
>> > >>> >
>> > >>> > > I can confirm that this segfaulting test has a big impact.
>> > >>> > >
>> > >>> > > On Wed, Jun 13, 2018 at 9:39 PM Aaron Markham
>> > >>> > > > > >>> > >
>> > >>> > > wrote:
>> > >>> > >
>> > >>> > > > I'd keep an eye on this one...  Flaky test:
>> > test_gru_bidirectional
>> > >>> > #11219
>> > >>> > > >
>> > >>> > > > https://github.com/apache/incubator-mxnet/issues/11219
>> > >>> > > >
>> > >>> > > > Just reran several PR's CI runs that all had the same
>> error!
>> > >>> > > >
>> > >>> > > > On Wed, Jun 13, 2018 at 5:42 PM, Anirudh <
>> anirudh2...@gmail.com>
>> > >>> > wrote:
>> > >>> > > >
>> > >>> > > > > Hi,
>> > >>> > > > >
>> > >>> > > > > PRs still in progress : #11127, #11236, #11210, #11054,
>> #11216.
>> > >>> > > > >
>> > >>> > > > > We are currently facing two issues which are delaying
>> the merge
>> > >>> > > > > of
>> > >>> > some
>> > >>> > > > of
>> > >>> > > > > these PRs:
>> > >>> > > > > 1. Flaky tests for scala API. A PR is already out to
>> disable
>> > the
>> > >>> > test:
>> > >>> > > > > https://github.com/apache/incubator-mxnet/issues/11249
>> > >>> > > > > 2. Builds breaking on windows:
>> > >>> > > > > https://github.com/apache/incubator-mxnet/issues/11265
>> > >>> > > > >
>> > >>> > > > > Anirudh
>> > >>> > > > >
>> > >>> > > > >
>> > >>> > > > > On Tue, Jun 12, 2018 at 11:59 AM, Anirudh
>> 

Re: users@mxnet

[YiZhi Liu]

I am personally not a big fan of mailing list but agree with Thomas
that we may get extra users, which worth a try.
On the other hand, I also have concern that we do not have a community
big enough to support multiple forums. If people asked questions but
got no response, that can be worse than not having the mailing list at
all.
On Mon, Jun 18, 2018 at 5:46 PM Thomas DELTEIL
 wrote:
>
> I was actually the one stating that we didn't need a user mailing list
> during the Seattle meetup, given all the reasons already exposed above.
>
> However given what proponents of a mailing list said, I personally wouldn't
> mind adding a new channel as a user mailing list, and monitoring it. There
> seems to be a subset of users, used to apache projects, that wouldn't use
> the forum but would use a mailing list. Though I think it is not as
> feature-rich as the forum and there is a risk of dilution of information.
> It is more about reaching those extra users. If we see a dilution of
> traffic on the forum towards the mailing list (~currently 100 posts/week)
> then maybe we can reconsider our assumptions?
>
> All the best,
>
> Thomas Delteil
>
> On Mon, Jun 18, 2018, 17:30 Pedro Larroy 
> wrote:
>
> > I agree with Tianqi, Eric and others. We shouldn't dilute the community
> > with another forum. Disqus is already working and has healthy
> > participation, you can get an email digest if you so desire. Subscribing to
> > a mailing list to get a question answered is quite a heavyweight investment
> > for many people and users who might not have the resources nor mental
> > bandwidth to receive more email volume in their inboxes.
> >
> > On Mon, Jun 18, 2018 at 10:19 AM Tianqi Chen 
> > wrote:
> >
> > > The problem of having multiple separate channels of communication is that
> > > users get confused, and the cost of maintenance goes up(people have to
> > > watch both). As the current community was at discuss forum and many users
> > > prefer it, having a mail-list is only a burden we will bring
> > >
> > > Tianqi
> > >
> > > On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski  wrote:
> > >
> > > > IMO, that is the wrong way to look at it.
> > > >
> > > > A users@ mailing list is a great, easy, low-cost and low-overhead way
> > of
> > > > *increasing* the user community and providing an extra level of
> > support.
> > > > Unless there is "strong evidence" that this is NOT the case, I would
> > > > recommend we create the list.
> > > >
> > > > > On Jun 16, 2018, at 12:28 AM, Tianqi Chen 
> > > > wrote:
> > > > >
> > > > > So unless there is a strong evidence that our community users prefers
> > > the
> > > > > mail-list, I would recommend we keep the current way
> > > > >
> > > > > Tianqi
> > > > >
> > > > > On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández  > >
> > > > wrote:
> > > > >
> > > > >> Are we targeting just Seattle as our community? I really hope we are
> > > > >> thinking a bit beyond that...
> > > > >>
> > > > >> On Fri, Jun 15, 2018, 21:22 Tianqi Chen 
> > > > wrote:
> > > > >>
> > > > >>> I remember last time during the mxnet meetup in Seattle, we did a
> > > > survey,
> > > > >>> and most users preferred the current discuss forum. So I would say
> > we
> > > > >> stick
> > > > >>> with that given the user community prefers that
> > > > >>>
> > > > >>> Tianqi
> > > > >>>
> > > > >>> On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández <
> > wik...@apache.org
> > > >
> > > > >>> wrote:
> > > > >>>
> > > >  Then, if everybody agree, let's request the mailing list creation
> > to
> > > > >>> INFRA
> > > >  ;-)
> > > > 
> > > >  Marco, I wouldn't do that. Typically developers are also
> > subscribed
> > > > >>> there,
> > > >  since they may be the most informed people for answering users'
> > > > >>> questions.
> > > >  But the topics discussed there may not be of the interest for pure
> > > >  development purposes. Some discussions will jump from users@ to
> > > dev@,
> > > > >>> but
> > > >  at a different level. So I wouldn't forward one mailing list to
> > the
> > > > >>> other.
> > > > 
> > > >  On Fri, Jun 15, 2018, 21:01 Marco de Abreu
> > > >   wrote:
> > > > 
> > > > > I think nobody was opposed to it in the past, right?
> > > > >
> > > > > I'd propose that all emails automatically get copied to dev@ to
> > > > >> ensure
> > > > > high
> > > > > visibility initially. What do you think?
> > > > >
> > > > > Sebastian  schrieb am Fr., 15. Juni 2018, 20:51:
> > > > >
> > > > >> I have already proposed this many times in the past and would
> > > > >>> strongly
> > > > >> encourage it.
> > > > >>
> > > > >> -s
> > > > >>
> > > > >> On 15.06.2018 21:56, Sergio Fernández wrote:
> > > > >>> Hi,
> > > > >>>
> > > > >>> is there any good reason why the podling doesn't have a users@
> > > >  mailing
> > > > >> list
> > > > >>> yet?
> > > > >>>
> > > > >>> Honestly speaking, I'm not a big 

Re: users@mxnet

[Thomas DELTEIL]

I was actually the one stating that we didn't need a user mailing list
during the Seattle meetup, given all the reasons already exposed above.

However given what proponents of a mailing list said, I personally wouldn't
mind adding a new channel as a user mailing list, and monitoring it. There
seems to be a subset of users, used to apache projects, that wouldn't use
the forum but would use a mailing list. Though I think it is not as
feature-rich as the forum and there is a risk of dilution of information.
It is more about reaching those extra users. If we see a dilution of
traffic on the forum towards the mailing list (~currently 100 posts/week)
then maybe we can reconsider our assumptions?

All the best,

Thomas Delteil

On Mon, Jun 18, 2018, 17:30 Pedro Larroy 
wrote:

> I agree with Tianqi, Eric and others. We shouldn't dilute the community
> with another forum. Disqus is already working and has healthy
> participation, you can get an email digest if you so desire. Subscribing to
> a mailing list to get a question answered is quite a heavyweight investment
> for many people and users who might not have the resources nor mental
> bandwidth to receive more email volume in their inboxes.
>
> On Mon, Jun 18, 2018 at 10:19 AM Tianqi Chen 
> wrote:
>
> > The problem of having multiple separate channels of communication is that
> > users get confused, and the cost of maintenance goes up(people have to
> > watch both). As the current community was at discuss forum and many users
> > prefer it, having a mail-list is only a burden we will bring
> >
> > Tianqi
> >
> > On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski  wrote:
> >
> > > IMO, that is the wrong way to look at it.
> > >
> > > A users@ mailing list is a great, easy, low-cost and low-overhead way
> of
> > > *increasing* the user community and providing an extra level of
> support.
> > > Unless there is "strong evidence" that this is NOT the case, I would
> > > recommend we create the list.
> > >
> > > > On Jun 16, 2018, at 12:28 AM, Tianqi Chen 
> > > wrote:
> > > >
> > > > So unless there is a strong evidence that our community users prefers
> > the
> > > > mail-list, I would recommend we keep the current way
> > > >
> > > > Tianqi
> > > >
> > > > On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández  >
> > > wrote:
> > > >
> > > >> Are we targeting just Seattle as our community? I really hope we are
> > > >> thinking a bit beyond that...
> > > >>
> > > >> On Fri, Jun 15, 2018, 21:22 Tianqi Chen 
> > > wrote:
> > > >>
> > > >>> I remember last time during the mxnet meetup in Seattle, we did a
> > > survey,
> > > >>> and most users preferred the current discuss forum. So I would say
> we
> > > >> stick
> > > >>> with that given the user community prefers that
> > > >>>
> > > >>> Tianqi
> > > >>>
> > > >>> On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández <
> wik...@apache.org
> > >
> > > >>> wrote:
> > > >>>
> > >  Then, if everybody agree, let's request the mailing list creation
> to
> > > >>> INFRA
> > >  ;-)
> > > 
> > >  Marco, I wouldn't do that. Typically developers are also
> subscribed
> > > >>> there,
> > >  since they may be the most informed people for answering users'
> > > >>> questions.
> > >  But the topics discussed there may not be of the interest for pure
> > >  development purposes. Some discussions will jump from users@ to
> > dev@,
> > > >>> but
> > >  at a different level. So I wouldn't forward one mailing list to
> the
> > > >>> other.
> > > 
> > >  On Fri, Jun 15, 2018, 21:01 Marco de Abreu
> > >   wrote:
> > > 
> > > > I think nobody was opposed to it in the past, right?
> > > >
> > > > I'd propose that all emails automatically get copied to dev@ to
> > > >> ensure
> > > > high
> > > > visibility initially. What do you think?
> > > >
> > > > Sebastian  schrieb am Fr., 15. Juni 2018, 20:51:
> > > >
> > > >> I have already proposed this many times in the past and would
> > > >>> strongly
> > > >> encourage it.
> > > >>
> > > >> -s
> > > >>
> > > >> On 15.06.2018 21:56, Sergio Fernández wrote:
> > > >>> Hi,
> > > >>>
> > > >>> is there any good reason why the podling doesn't have a users@
> > >  mailing
> > > >> list
> > > >>> yet?
> > > >>>
> > > >>> Honestly speaking, I'm not a big fan of the other tools the
> > > >> podling
> > >  is
> > > >>> using. Slack and Web forums a cool tools, and I used them a lot
> > > >> in
> > > > other
> > > >>> contexts. But when it comes to transparency and community,
> > > >> mailing
> > > > lists
> > > >>> play a crucial role in the Apache Way.
> > > >>>
> > > >>> Users are the most important asset a project can have. Even
> more
> > > >>> than
> > > >>> developers, believe me. So I think it's time to create a users@
> > > > mailing
> > > >>> list for to helping MXNet grow its community beyong the core
> > > >> team.
> > > >>>

Re: users@mxnet

[Pedro Larroy]

I agree with Tianqi, Eric and others. We shouldn't dilute the community
with another forum. Disqus is already working and has healthy
participation, you can get an email digest if you so desire. Subscribing to
a mailing list to get a question answered is quite a heavyweight investment
for many people and users who might not have the resources nor mental
bandwidth to receive more email volume in their inboxes.

On Mon, Jun 18, 2018 at 10:19 AM Tianqi Chen 
wrote:

> The problem of having multiple separate channels of communication is that
> users get confused, and the cost of maintenance goes up(people have to
> watch both). As the current community was at discuss forum and many users
> prefer it, having a mail-list is only a burden we will bring
>
> Tianqi
>
> On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski  wrote:
>
> > IMO, that is the wrong way to look at it.
> >
> > A users@ mailing list is a great, easy, low-cost and low-overhead way of
> > *increasing* the user community and providing an extra level of support.
> > Unless there is "strong evidence" that this is NOT the case, I would
> > recommend we create the list.
> >
> > > On Jun 16, 2018, at 12:28 AM, Tianqi Chen 
> > wrote:
> > >
> > > So unless there is a strong evidence that our community users prefers
> the
> > > mail-list, I would recommend we keep the current way
> > >
> > > Tianqi
> > >
> > > On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández 
> > wrote:
> > >
> > >> Are we targeting just Seattle as our community? I really hope we are
> > >> thinking a bit beyond that...
> > >>
> > >> On Fri, Jun 15, 2018, 21:22 Tianqi Chen 
> > wrote:
> > >>
> > >>> I remember last time during the mxnet meetup in Seattle, we did a
> > survey,
> > >>> and most users preferred the current discuss forum. So I would say we
> > >> stick
> > >>> with that given the user community prefers that
> > >>>
> > >>> Tianqi
> > >>>
> > >>> On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández  >
> > >>> wrote:
> > >>>
> >  Then, if everybody agree, let's request the mailing list creation to
> > >>> INFRA
> >  ;-)
> > 
> >  Marco, I wouldn't do that. Typically developers are also subscribed
> > >>> there,
> >  since they may be the most informed people for answering users'
> > >>> questions.
> >  But the topics discussed there may not be of the interest for pure
> >  development purposes. Some discussions will jump from users@ to
> dev@,
> > >>> but
> >  at a different level. So I wouldn't forward one mailing list to the
> > >>> other.
> > 
> >  On Fri, Jun 15, 2018, 21:01 Marco de Abreu
> >   wrote:
> > 
> > > I think nobody was opposed to it in the past, right?
> > >
> > > I'd propose that all emails automatically get copied to dev@ to
> > >> ensure
> > > high
> > > visibility initially. What do you think?
> > >
> > > Sebastian  schrieb am Fr., 15. Juni 2018, 20:51:
> > >
> > >> I have already proposed this many times in the past and would
> > >>> strongly
> > >> encourage it.
> > >>
> > >> -s
> > >>
> > >> On 15.06.2018 21:56, Sergio Fernández wrote:
> > >>> Hi,
> > >>>
> > >>> is there any good reason why the podling doesn't have a users@
> >  mailing
> > >> list
> > >>> yet?
> > >>>
> > >>> Honestly speaking, I'm not a big fan of the other tools the
> > >> podling
> >  is
> > >>> using. Slack and Web forums a cool tools, and I used them a lot
> > >> in
> > > other
> > >>> contexts. But when it comes to transparency and community,
> > >> mailing
> > > lists
> > >>> play a crucial role in the Apache Way.
> > >>>
> > >>> Users are the most important asset a project can have. Even more
> > >>> than
> > >>> developers, believe me. So I think it's time to create a users@
> > > mailing
> > >>> list for to helping MXNet grow its community beyong the core
> > >> team.
> > >>>
> > >>> Cheers,
> > >>>
> > >>
> > >
> > 
> > >>>
> > >>
> >
> >
>

Re: users@mxnet

[Jim Jagielski]

"Mailing list is an obsolete legacy for old projects"...

Not true. If you don't understand the reason and rationale and *benefits* of 
using a mailing list, and why they are so core to how Apache runs projects, I'd 
be happy to provide some data.

> On Jun 18, 2018, at 2:53 PM, Eric Xie  wrote:
> 
> Neither TF nor Pytorch uses mailing lists though. In fact I can't think of 
> any deep learning community that uses mailing lists. Mailing list is an 
> obsolete legacy for old projects. No point in bringing it into new projects.
> 
> Thanks,
> Eric
> 
> On 2018/06/18 18:42:12, Sebastian  wrote: 
>> I am puzzled by the reluctance of this project to setup a user 
>> mailinglist to be honest.
>> 
>> MXNet has major issues with attracting a community outside of Amazon 
>> (whenever I hear folks talking about deep learning, they usually mention 
>> tensorflow, pytorch and keras, but I rarely hear someone talk about 
>> MXNet). At the same time, there is so much resistance to adopt practices 
>> that are successfully used by many high-profile toplevel projects...
>> 
>> -s
>> 
>> On 18.06.2018 20:37, Timur Shenkao wrote:
>>> Facebook is definitely a bad idea: we will be dependent on third party
>>> provider + unclear who & how manages such group etc.
>>> Forum + Confluence + Slack is  much better then.
>>> 
>>> On Mon, Jun 18, 2018 at 7:17 PM, Ivan Serdyuk 
>>> wrote:
>>> 
 Greetings Barber, Christopher. I had an idea to move out some discussions,
 covering Java and Scala API, to Facebook. So if somewhere exists a local
 JUG or Scala user group - they could reflect the topic of discussion. But
 background stuff could take place on mailing lists, Slack, forum, whatever.
 The reverse mechanism could be used to involve new committers, as well (so
 they would appear as presented newcomers, as for contributions).
 
 Regards
 Ivan
 
 On Mon, Jun 18, 2018 at 8:40 PM, Barber, Christopher <
 christopher.bar...@analog.com> wrote:
 
> I don't understand why you would want a users mailing list when you
> already have discussion forums. Users that want to be notified of new
 posts
> on the forum can configure their notification preferences appropriately.
> The traffic on the forums is already pretty low. I would think you would
> not want to dilute that further.
> 
> Christopher
> 
> On 6/18/18, 1:27 PM, "Jim Jagielski"  wrote:
> 
> users@ mailing lists have great societal advantages that one
> shouldn't ignore...
> 
> And it's not like this is the only project with "multiple"
> communication choices for users. Most, if not all, projects have users@in
> addition to such supplemental methods as IRC channels, a forum, etc...
 It's
> about making it easy to have as many users as possible and as many
> potential ways for users to communicate. It's not confusing; it's
> empowering :)
> 
>> On Jun 18, 2018, at 1:19 PM, Tianqi Chen  
> wrote:
>> 
>> The problem of having multiple separate channels of communication
 is
> that
>> users get confused, and the cost of maintenance goes up(people have
> to
>> watch both). As the current community was at discuss forum and many
> users
>> prefer it, having a mail-list is only a burden we will bring
>> 
>> Tianqi
>> 
>> On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski 
> wrote:
>> 
>>> IMO, that is the wrong way to look at it.
>>> 
>>> A users@ mailing list is a great, easy, low-cost and low-overhead
> way of
>>> *increasing* the user community and providing an extra level of
> support.
>>> Unless there is "strong evidence" that this is NOT the case, I
 would
>>> recommend we create the list.
>>> 
 On Jun 16, 2018, at 12:28 AM, Tianqi Chen <
> tqc...@cs.washington.edu>
>>> wrote:
 
 So unless there is a strong evidence that our community users
> prefers the
 mail-list, I would recommend we keep the current way
 
 Tianqi
 
 On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández <
> wik...@apache.org>
>>> wrote:
 
> Are we targeting just Seattle as our community? I really hope we
> are
> thinking a bit beyond that...
> 
> On Fri, Jun 15, 2018, 21:22 Tianqi Chen <
 tqc...@cs.washington.edu
>> 
>>> wrote:
> 
>> I remember last time during the mxnet meetup in Seattle, we
 did a
>>> survey,
>> and most users preferred the current discuss forum. So I would
> say we
> stick
>> with that given the user community prefers that
>> 
>> Tianqi
>> 
>> On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández <
> wik...@apache.org>
>> wrote:
>> 
>>> Then, if everybody agree, let's request the mailing list
> creation t

Additional mentor to MXNet - Jim Jagielski

[Naveen Swamy]

Hi All,

I am excited to announce that we have an additional mentor for MXNet: Jim
Jagielski, the cofounder of Apache, he brings vast experience in building
and growing successful communities around projects. I am sure we will be
able to tap into his experience to bring alignment and build a strong
community around MXNet.

He has a page on wikipedia, check it out :)
https://en.wikipedia.org/wiki/Jim_Jagielski.

Thanks, Naveen

FOSS BackStage/Apache EU Roadshow summary

[Naveen Swamy]

Hi all,

I was at the FOSS Backstage/Apache EU Roadshow last week, it covered many
important topics in open-source software and governance. I also gave a
presentation on Distributed Inference using Apache MXNet and Apache Spark.

I also had the opportunity to listen and meet some of the Open source
pioneers and discuss various aspects of the project and seek their guidance
on how to build successful community around MXNet. Talking to board members
gave several insights that could be helpful in growing our community such
as having meetings at different time zones, etc.,

Here is the summary of the various sessions I attended and I hope you'll
find it useful.

*Keynote by Danese Cooper : *

She is well known for being instrumental in open-sourcing Java while at Sun
Microsystems. Danese shared her success stories from her 20 years of
Experience at Open source and key elements necessary for open source success

   - Transparency is not negotiable in Open source, it not enough to ride
   the movement but also important to shape. Called upon each one of us to
   review Open-source history, Apache way and help shape the
   - Open source is all about people, At Aapche we say Community over Code
   — It is important to speak the truth, it even more important you are kind
   in telling the truth.
   - Companies want to use it for strategic advantage:  She shared the
   story of Sun wanting to open source Java since 1999 but there was no
   alignment amongst the leaders and it happened only in 2006 due to largely
   due ASF telling them it was going to leave the effort which made Sun to
   change their mind and come up with the Apache Compromise
   
   .
   - Develop Patronage and shared the story of how GNOME open source
   project had world class accessibility tools that Sun was able to use.
   - Money changes everything even open source
   - Focus on Tech: Focus on building a differentiator for your project
   that users would love and want to come to you.



*They don’t understand me — Tales from Multicultural trenches.*

Bertrand Delacrétaz, Director of Apache and Principal Scientist

   - Cultural diversity is a great asset in our community but makes it
   harder to communicate especially on low-bandwidth channels such as email.
   - Culture does not necessarily mean only origin, it is your affiliation
   to sports, Education, various facets to life.
   - When you work on a project led by a community, you count on your
   fellow contributors to course correct your blindsight -- involve them by
   being friendly and welcoming
   - Communication is really complicated and cultures hides it in
   unexpected places.
   - To avoid misunderstanding, use simple language.
   - Avoid misunderstanding by assuming good intentions, ask for
   clarification
   - Reformulate your understanding without being aggressive, very often
   the message is not what you understood.
   - For the sake of the project find ways to work.
   - Mistakes happen, don’t be hung up on it and instead find the course
   correction.
   - talk less. Code speaks louder than words! — show them the code -
   prototype, unfinished, ugly it doesn’t matter
   - Find allies on the other side of the aisle and work with them and make
   them role models, people tend to trust of the same clan.



*Apache Flink:*

Fabian Heuske – Cofounder of Apache Flink & Data Artisans.

   - Apache Flink focuses on growing community in a competitive landscape.
   - Speaking at Conferences and Meetups helped spread the word about Flink.
   - *Find early users/partners and growing them into evangelists also
   helped in its growth*
   - Find your niche — Stream processing though was hard against biggest
   players on their turf.
   - Being just better at the same thing does not cut it.
   - Community and Minshare - technology is easier to evolve than
   community,
   - In 2016, Flink pioneers many important stream processing and added
   features that no else had.
   - Being in Apache was very good to get mentorship in building the
   project and get legal protection.
   - Mentors associated during incubation helped a lot in growing the
   community.
   - Community trumps technology to a large extent.
   - *Being the first of its kind is also important, unique technology also
   attracts community.*


*The Apache Way - Panel discussion*

*If it didn’t happen on the list, it didn’t happen*

   - works across timezones
   - being transparent helps when you want to call for help since you have
   built the trust.
   - being asynchrnous helps for people who are really volunteering and can
   find time at night or weekends.
   - can go back in time and look why we did it in a certain way.
   - easy to follow for users, they cannot become active if you don't share.

*Minimum of 72 hours.*

   - helps contributors across different timezones.

*Voting*

   - Vote in mo

Re: Config for Scala Maven Package

[Naveen Swamy]

Anirudh, I have tried to fix and automate parts of the publish process and
written down the process. If everything works as expected & since it takes
building on 3 different platforms it could take 4-6 hours. We will also
continue to automate this further. This is excluding the time it needs to
fix various bugs if any.

On Mon, Jun 18, 2018 at 11:39 AM, Anirudh  wrote:

> Hi Naveen,
>
> Sorry I missed this. Can you please elaborate on how long it takes
> typically for publishing the scala packages to maven before the vote?
> I think if there is considerable churn in publishing these packages today,
> we should not wait for the package for the vote as we may sacrifice our big
> python user base
> getting the release early for the scala user base.
>
> Anirudh
>
>
> On Thu, May 31, 2018 at 7:27 AM, Naveen Swamy  wrote:
>
> > All, I have created a guide to maven publish process here;
> > https://cwiki.apache.org/confluence/display/MXNET/
> > MXNet-Scala+Release+Process
> > .
> > Going forward, I propose that we include the scala maven packages also in
> > our RC voting(I have updated the release-process already), I understand
> the
> > maven packages use the same code base that is sent for voting. I
> > encountered a quite bit of issues trying to publish the package, this
> could
> > because we never did this past and I have automated as much as possible.
> I
> > think there is value in getting these packages tested during the RC
> > process.
> >
> > Sandeep,
> > I am using what is available on AWS Deep Learning Base AMI which is 3.4
> and
> > CUDNN 7.0.5.
> >
> > Thanks, Naveen
> >
> > On Wed, May 23, 2018 at 8:57 AM, sandeep krishnamurthy <
> > sandeep.krishn...@gmail.com> wrote:
> >
> > > Hi Naveen,
> > >
> > > Great work.
> > > I was wondering about OpenCV and cuDNN
> > >
> > > Thanks,
> > > Sandeep
> > >
> > > On Mon, May 21, 2018 at 5:44 PM, Naveen Swamy 
> > wrote:
> > >
> > > > Hi Marco,
> > > > For MKL/MKLDNN --> I don't think we should build since its still
> > > > experimental. Its sad that MKLML was removed without a proper
> > discussion,
> > > > granted
> > > > I have not tested with CUDA 9.2, I will building on a AWS Ubuntu
> > instance
> > > > that is built with CUDA 9.0, I presume it should work fine.
> > > >
> > > >
> > > > On Mon, May 21, 2018 at 5:31 PM, Marco de Abreu <
> > > > marco.g.ab...@googlemail.com> wrote:
> > > >
> > > > > Hi Naveen,
> > > > >
> > > > > thank you for driving the releases for the Scala packages!
> > > > >
> > > > > For CPU, I'd propose add MKL/MKLDNN builds.
> > > > >
> > > > > For GPU, do we have to specify the minor version of CUDA up front
> or
> > > are
> > > > > they interchangeable? CUDA 9.2, for example, received some quite
> > > > > significant performance updates and it would be great if our users
> > > could
> > > > > use them.
> > > > >
> > > > > Best regards,
> > > > > Marco
> > > > >
> > > > > On Fri, May 18, 2018 at 8:28 PM, Naveen Swamy 
> > > > wrote:
> > > > >
> > > > > > I am working on publishing the Scala package from 1.2.0 release.
> > What
> > > > do
> > > > > > you recommend should be the configuration for the Scala Packages
> ?
> > > > > >
> > > > > > I am thinking of the below
> > > > > > For CPU build with
> > > > > > OPENBLAS
> > > > > > LAPACK
> > > > > >
> > > > > > For GPU build with
> > > > > > OPENBLAS
> > > > > > CUDA 9.0
> > > > > > NCCL
> > > > > >
> > > > > > Let me know if you concerns or recommendations.
> > > > > >
> > > > > > Thanks, Naveen
> > > > > >
> > > > >
> > > >
> > >
> > >
> > >
> > > --
> > > Sandeep Krishnamurthy
> > >
> >
>

Re: Nightly tests README accurate?

[Indhu]

I was working on testing the notebooks from https://gluon.mxnet.io/ as part
of the nightly tests to make sure we notice any API change that could break
code in those notebooks.

The README in the nightly folder said the tests are being run on machines
with multiple GPUs. I was wondering if I have to run multiple notebooks in
parallel to efficiently utilize all GPUs. After reading your email, I'm
thinking I'll run most of the notebooks on P3.2xl and only the ones that
require multiple GPUs on P3.8xl. Thanks for the reply.


On Mon, Jun 18, 2018 at 2:24 PM Marco de Abreu
 wrote:

> In future (next few days) the nightlies will be running on our regular CI
> at http://jenkins.mxnet-ci.amazon-ml.com/
>
> On there, we support C5.18xlarge (Ubuntu, CentOS within Docker and
> Windows), G3.8xlarge (Ubuntu, CentOS within Docker and Windows), P3.2xlarge
> (Ubuntu, CentOS within docker) and (upcoming) P3.8xlarge (Ubuntu, CentOS
> within docker).
>
> What exact requirements do you have, Indhu? I'm happy to assist!
>
> -Marco
>
> On Mon, Jun 18, 2018 at 1:47 PM Indhu  wrote:
>
> > Thanks Meghna. Do you know what hardware the nightly tests are currently
> > running on? I'm adding some nightly tests. It will be useful to know.
> >
> >
> > On Mon, Jun 18, 2018 at 9:11 AM Meghna Baijal <
> meghnabaijal2...@gmail.com>
> > wrote:
> >
> > > Hi,
> > > This readme is incorrect and I am fixing this as a part of the nightly
> > > tests PR soon.
> > >
> > > Thanks,
> > > Meghna
> > >
> > > On Sun, Jun 17, 2018 at 7:05 AM kellen sunderland <
> > > kellen.sunderl...@gmail.com> wrote:
> > >
> > > > I'm not a commiter, but I would say this is clearly not correct.  I
> > would
> > > > propose removing the doc until someone has time to verify that the
> > steps
> > > > work correctly, and that the descriptions are accurate.
> > > >
> > > > On Thu, Jun 14, 2018 at 10:04 PM Indhu 
> > wrote:
> > > >
> > > > > Is the README
> > > > > <
> https://github.com/apache/incubator-mxnet/tree/master/tests/nightly
> > >
> > > > for
> > > > > the nightly tests accurate? For example,
> > > > >
> > > > > 1. Are tests being run on machines with Intel i7-4790 and 4 Nvidia
> > GTX
> > > > 970
> > > > > Tis?
> > > > > 2. Is http://ci.dmlc.ml/ the right place to look for build status?
> > > > > 3. Is the instruction to run on Jenkins correct?
> > > > >
> > > > > If not, what all needs to be changed in that page?
> > > > >
> > > >
> > >
> >
>

Single-Machine Topology-aware Communication

[Carl Yang]

Hi,

Currently, we have two methods for single-machine communication:
parameter server and NCCL ring reduction. Both of these methods have
some downsides. Parameter server does not differentiate between NVLink
connections and PCI-E, so it ends up using the higher latency and
slower PCI-E connections as frequently as it does NVLink. NCCL uses
the ring reduce algorithm, which has higher theoretical latency than
other algorithms. I am working on a topology-aware approach that can
address these limitations. Design proposal is on cwiki:
https://cwiki.apache.org/confluence/display/MXNET/Single+machine+All+Reduce+Topology-aware+Communication

Please feel free to let me know if you have any suggestions.

Regards,
Carl

Re: Nightly tests README accurate?

[Marco de Abreu]

In future (next few days) the nightlies will be running on our regular CI
at http://jenkins.mxnet-ci.amazon-ml.com/

On there, we support C5.18xlarge (Ubuntu, CentOS within Docker and
Windows), G3.8xlarge (Ubuntu, CentOS within Docker and Windows), P3.2xlarge
(Ubuntu, CentOS within docker) and (upcoming) P3.8xlarge (Ubuntu, CentOS
within docker).

What exact requirements do you have, Indhu? I'm happy to assist!

-Marco

On Mon, Jun 18, 2018 at 1:47 PM Indhu  wrote:

> Thanks Meghna. Do you know what hardware the nightly tests are currently
> running on? I'm adding some nightly tests. It will be useful to know.
>
>
> On Mon, Jun 18, 2018 at 9:11 AM Meghna Baijal 
> wrote:
>
> > Hi,
> > This readme is incorrect and I am fixing this as a part of the nightly
> > tests PR soon.
> >
> > Thanks,
> > Meghna
> >
> > On Sun, Jun 17, 2018 at 7:05 AM kellen sunderland <
> > kellen.sunderl...@gmail.com> wrote:
> >
> > > I'm not a commiter, but I would say this is clearly not correct.  I
> would
> > > propose removing the doc until someone has time to verify that the
> steps
> > > work correctly, and that the descriptions are accurate.
> > >
> > > On Thu, Jun 14, 2018 at 10:04 PM Indhu 
> wrote:
> > >
> > > > Is the README
> > > >  >
> > > for
> > > > the nightly tests accurate? For example,
> > > >
> > > > 1. Are tests being run on machines with Intel i7-4790 and 4 Nvidia
> GTX
> > > 970
> > > > Tis?
> > > > 2. Is http://ci.dmlc.ml/ the right place to look for build status?
> > > > 3. Is the instruction to run on Jenkins correct?
> > > >
> > > > If not, what all needs to be changed in that page?
> > > >
> > >
> >
>

Re: Nightly tests README accurate?

[Indhu]

Thanks Meghna. Do you know what hardware the nightly tests are currently
running on? I'm adding some nightly tests. It will be useful to know.


On Mon, Jun 18, 2018 at 9:11 AM Meghna Baijal 
wrote:

> Hi,
> This readme is incorrect and I am fixing this as a part of the nightly
> tests PR soon.
>
> Thanks,
> Meghna
>
> On Sun, Jun 17, 2018 at 7:05 AM kellen sunderland <
> kellen.sunderl...@gmail.com> wrote:
>
> > I'm not a commiter, but I would say this is clearly not correct.  I would
> > propose removing the doc until someone has time to verify that the steps
> > work correctly, and that the descriptions are accurate.
> >
> > On Thu, Jun 14, 2018 at 10:04 PM Indhu  wrote:
> >
> > > Is the README
> > > 
> > for
> > > the nightly tests accurate? For example,
> > >
> > > 1. Are tests being run on machines with Intel i7-4790 and 4 Nvidia GTX
> > 970
> > > Tis?
> > > 2. Is http://ci.dmlc.ml/ the right place to look for build status?
> > > 3. Is the instruction to run on Jenkins correct?
> > >
> > > If not, what all needs to be changed in that page?
> > >
> >
>

Re: Clojure Package

[Carin Meier]

Hagay,

Thanks! I added the MXNet 1.3 page.

Please let me know if you need me to add any other details.

- Carin

On Mon, Jun 18, 2018 at 4:01 PM, Hagay Lupesko  wrote:

> Carin,
>
> Thanks for contributing so much to MXNet!
> I just went through the planned MXNet v1.3 release scope [1], and did not
> see the Clojure package. I think it would be a great addition to MXNet, and
> initially can be marked as experimental.
>
> Did you consider adding it to MXNet 1.3 as an experimental feature?
>
> Hagay
>
> [1]
> https://cwiki.apache.org/confluence/display/MXNET/
> Project+Proposals+for+next+MXNet+Release
>
>
> On Fri, Jun 15, 2018 at 5:52 AM Carin Meier  wrote:
>
> > Kovas Boguta https://twitter.com/kovasb, from the Clojure/AI community,
> > graciously took some time to review the PR for the clojure package.
> >
> > He had some insightful feedback and high level questions that I thought
> > might be of interest to the larger dev mailing list.
> >
> > https://github.com/apache/incubator-mxnet/pull/11205
> >
> > Feel free to join in on the discussion here or on the PR.
> >
> > - Carin
> >
> > On Mon, Jun 11, 2018 at 6:49 PM, Carin Meier 
> wrote:
> >
> > > I'm fine with whatever process works best and what makes everyone the
> > most
> > > comfortable.
> > >
> > > I've completed one of the requests for code coverage and integrated the
> > > cloverage, (https://github.com/cloverage/cloverage), plugin in the PR
> > > branch. I've published the results to the confluence page to help with
> > > transparency and give everyone a better idea where the level of testing
> > is
> > > currently.
> > > https://cwiki.apache.org/confluence/display/MXNET/
> > > Clojure+Package+Contribution+Needs
> > >
> > >
> > >
> > > On Mon, Jun 11, 2018 at 6:24 PM, Marco de Abreu <
> > > marco.g.ab...@googlemail.com> wrote:
> > >
> > >> Exactly, that's what I'm proposing. Having the migration in multiple
> > >> separate PRs which are done in sequence. This would mean that the
> > initial
> > >> PRs might not be tested.
> > >> We could make the factors you mentioned acceptance criteria to be
> moved
> > >> out
> > >> of contrib.
> > >>
> > >> -Marco
> > >>
> > >> On Mon, Jun 11, 2018 at 3:17 PM Naveen Swamy 
> > wrote:
> > >>
> > >> > I did not suggest to do it at once. I am not comfortable to merge
> code
> > >> > without good tests, documentation and examples that is not to say
> > >> Clojure
> > >> > codebase does not have that.
> > >> > All that you are saying can happen in separate PRs if you want to
> > break
> > >> it
> > >> > up.
> > >> >
> > >> >
> > >> > On Tue, Jun 12, 2018 at 12:07 AM, Marco de Abreu <
> > >> > marco.g.ab...@googlemail.com> wrote:
> > >> >
> > >> > > The problem I see here is that the migration will have different
> > type
> > >> of
> > >> > > challenges which should be handled isolated. Trying to solve them
> > all
> > >> at
> > >> > > once will make it very lengthy and also hard to review.
> Considering
> > >> Carin
> > >> > > and her team this is doing this on a voluntary base, I'd like to
> > keep
> > >> the
> > >> > > number of hoops to jump through per stage as small as possible and
> > >> rather
> > >> > > split it up into multiple efforts.
> > >> > >
> > >> > > If we would do everything at once, we'd have to involve a lot of
> > >> people
> > >> > and
> > >> > > it would be hard to review. We'd need at least two or three
> > reviewers
> > >> > > involved in that process: You (or another committer familiar with
> > >> Scala
> > >> > to
> > >> > > review the Scala part), Yi Zhi (general reviewer) and me (CI
> > >> > integration).
> > >> > > It would probably require even more committers for other stuff
> that
> > >> comes
> > >> > > up. It would rather be better to keep the parts that have to be
> > >> touched
> > >> > as
> > >> > > isolated and few as possible.
> > >> > > For example, after the code has been approved and merged in
> > general, I
> > >> > can
> > >> > > assist with the CI integration. This would not require oversight
> > from
> > >> > other
> > >> > > committers, so they'd be free. After that, we'd need somebody
> > familiar
> > >> > with
> > >> > > the release process (probably Naveen) and I'd be free after that.
> > >> Then we
> > >> > > need general improvements which would also involve other people
> > again.
> > >> > > Trying to squeeze everything into a single stage is going to make
> it
> > >> very
> > >> > > hard in my opinion.
> > >> > >
> > >> > >
> > >> > > -Marco
> > >> > >
> > >> > > On Mon, Jun 11, 2018 at 2:56 PM Naveen Swamy 
> > >> wrote:
> > >> > >
> > >> > > > I disagree with your approach, We should bring features
> > iteratively
> > >> > well
> > >> > > > tested and well documented. MXNet already has many different
> > >> language
> > >> > > > bindings which has quite a bit of tech-debt, I don't want just
> to
> > >> add
> > >> > > more
> > >> > > > tech-debt to the code base with new language bindings as well.
> > >> > > >
> > >> > > > On Mon, Jun 11, 2018 at 11:17 PM

Re: Update on 1.2.1 release

[Anirudh]

Hi Sina,

RC is available here:
https://github.com/apache/incubator-mxnet/releases/tag/1.2.1.rc0 . We are
waiting to publish scala package to maven before we start the vote. Naveen,
Qing and Andrew are working on this.

Anirudh

On Mon, Jun 18, 2018 at 1:18 PM, Afrooze, Sina 
wrote:

> Just wondering if there is any estimate on when a 1.2.1 RC may be created?
> We are using 1.3.0 beta builds because those are the only pip-install-able
> builds available, but they introduce some instability in our workflow that
> we'd like to avoid by switching to a 1.2.1 build. - Sina
>
> On 6/15/18, 12:55 PM, "Marco de Abreu" 
> wrote:
>
> Hello,
>
> we would need this PR to be merged as well:
> https://github.com/apache/incubator-mxnet/pull/11309
>
> With the next deployment of new Windows slaves (which should happen
> asap),
> the 1.2 branch and all past branches are going to break since the test
> execution is not versioned. Please merge this PR to ensure we're still
> able
> to test the 1.2 branch on our CI.
>
> -Marco
>
> On Thu, Jun 14, 2018 at 6:20 PM Anirudh  wrote:
>
> > Hi all,
> >
> > We have one last PR before code freeze:
> > https://github.com/apache/incubator-mxnet/pull/11298
> >
> > Anirudh
> >
> > On Thu, Jun 14, 2018 at 11:46 AM, Anirudh 
> wrote:
> >
> > > Waiting on CI for the PRs: #11236, #11210, #11267
> > >
> > > Other PRs have been merged.
> > >
> > > On Wed, Jun 13, 2018 at 10:50 PM, Anirudh 
> wrote:
> > >
> > >> Thanks Tao! Yes, this shouldn't be a blocker for 1.2.1.
> > >>
> > >> On Wed, Jun 13, 2018 at 10:46 PM, Lv, Tao A 
> wrote:
> > >>
> > >>>
> > >>> Yes, #10311 is only in master branch, so I guess it won't impact
> 1.2.0
> > >>> branch and block the release of 1.2.1, right?
> > >>>
> > >>> A PR (#11273) is submitted to disable the test temporally and
> hopefully
> > >>> it will be fixed soon.
> > >>>
> > >>> -tao
> > >>>
> > >>> -Original Message-
> > >>> From: Marco de Abreu [mailto:marco.g.abreu@
> googlemail.com.INVALID]
> > >>> Sent: Thursday, June 14, 2018 1:21 PM
> > >>> To: dev@mxnet.incubator.apache.org
> > >>> Subject: Re: Update on 1.2.1 release
> > >>>
> > >>> On windows tests a segfault is indicated by "If -764728474728".
> I have
> > >>> also seen it happen on Ubuntu, there are probably some links in
> the
> > issue
> > >>> (on my phone right now).
> > >>>
> > >>> Anirudh  schrieb am Mi., 13. Juni 2018,
> 22:16:
> > >>>
> > >>> > By segfaulting test do you mean : test_gru_bidirectional. I
> don't see
> > >>> > the segfault in the logs. Can you point me to the test.
> > >>> > Also, this seems to be specific to the master and not in 1.2:
> > >>> > https://github.com/apache/incubator-mxnet/pull/10311
> > >>> >
> > >>> > Anirudh
> > >>> >
> > >>> > On Wed, Jun 13, 2018 at 10:00 PM, Marco de Abreu <
> > >>> > marco.g.ab...@googlemail.com.invalid> wrote:
> > >>> >
> > >>> > > I can confirm that this segfaulting test has a big impact.
> > >>> > >
> > >>> > > On Wed, Jun 13, 2018 at 9:39 PM Aaron Markham
> > >>> > >  > >>> > >
> > >>> > > wrote:
> > >>> > >
> > >>> > > > I'd keep an eye on this one...  Flaky test:
> > test_gru_bidirectional
> > >>> > #11219
> > >>> > > >
> > >>> > > > https://github.com/apache/incubator-mxnet/issues/11219
> > >>> > > >
> > >>> > > > Just reran several PR's CI runs that all had the same
> error!
> > >>> > > >
> > >>> > > > On Wed, Jun 13, 2018 at 5:42 PM, Anirudh <
> anirudh2...@gmail.com>
> > >>> > wrote:
> > >>> > > >
> > >>> > > > > Hi,
> > >>> > > > >
> > >>> > > > > PRs still in progress : #11127, #11236, #11210, #11054,
> #11216.
> > >>> > > > >
> > >>> > > > > We are currently facing two issues which are delaying
> the merge
> > >>> > > > > of
> > >>> > some
> > >>> > > > of
> > >>> > > > > these PRs:
> > >>> > > > > 1. Flaky tests for scala API. A PR is already out to
> disable
> > the
> > >>> > test:
> > >>> > > > > https://github.com/apache/incubator-mxnet/issues/11249
> > >>> > > > > 2. Builds breaking on windows:
> > >>> > > > > https://github.com/apache/incubator-mxnet/issues/11265
> > >>> > > > >
> > >>> > > > > Anirudh
> > >>> > > > >
> > >>> > > > >
> > >>> > > > > On Tue, Jun 12, 2018 at 11:59 AM, Anirudh
> > >>> > > > > 
> > >>> > > wrote:
> > >>> > > > >
> > >>> > > > > > Hi all,
> > >>> > > > > >
> > >>> > > > > > Here are the PRs that are being tracked for 1.2.1
> release:
> > >>> > > > > >
> > >>> > > > > > Related to the save_params backwards incompatible
> change:
> > >>> > > > > > #11127
> > >>> > (In
> > >>> > > > > > Progress), #11236 (In Progress), #11210 (In Progress)
> MKLDNN
> > >

Re: Update on 1.2.1 release

[Afrooze, Sina]

Just wondering if there is any estimate on when a 1.2.1 RC may be created? We 
are using 1.3.0 beta builds because those are the only pip-install-able builds 
available, but they introduce some instability in our workflow that we'd like 
to avoid by switching to a 1.2.1 build. - Sina 

On 6/15/18, 12:55 PM, "Marco de Abreu"  
wrote:

Hello,

we would need this PR to be merged as well:
https://github.com/apache/incubator-mxnet/pull/11309

With the next deployment of new Windows slaves (which should happen asap),
the 1.2 branch and all past branches are going to break since the test
execution is not versioned. Please merge this PR to ensure we're still able
to test the 1.2 branch on our CI.

-Marco

On Thu, Jun 14, 2018 at 6:20 PM Anirudh  wrote:

> Hi all,
>
> We have one last PR before code freeze:
> https://github.com/apache/incubator-mxnet/pull/11298
>
> Anirudh
>
> On Thu, Jun 14, 2018 at 11:46 AM, Anirudh  wrote:
>
> > Waiting on CI for the PRs: #11236, #11210, #11267
> >
> > Other PRs have been merged.
> >
> > On Wed, Jun 13, 2018 at 10:50 PM, Anirudh  wrote:
> >
> >> Thanks Tao! Yes, this shouldn't be a blocker for 1.2.1.
> >>
> >> On Wed, Jun 13, 2018 at 10:46 PM, Lv, Tao A  wrote:
> >>
> >>>
> >>> Yes, #10311 is only in master branch, so I guess it won't impact 1.2.0
> >>> branch and block the release of 1.2.1, right?
> >>>
> >>> A PR (#11273) is submitted to disable the test temporally and 
hopefully
> >>> it will be fixed soon.
> >>>
> >>> -tao
> >>>
> >>> -Original Message-
> >>> From: Marco de Abreu [mailto:marco.g.ab...@googlemail.com.INVALID]
> >>> Sent: Thursday, June 14, 2018 1:21 PM
> >>> To: dev@mxnet.incubator.apache.org
> >>> Subject: Re: Update on 1.2.1 release
> >>>
> >>> On windows tests a segfault is indicated by "If -764728474728". I have
> >>> also seen it happen on Ubuntu, there are probably some links in the
> issue
> >>> (on my phone right now).
> >>>
> >>> Anirudh  schrieb am Mi., 13. Juni 2018, 22:16:
> >>>
> >>> > By segfaulting test do you mean : test_gru_bidirectional. I don't 
see
> >>> > the segfault in the logs. Can you point me to the test.
> >>> > Also, this seems to be specific to the master and not in 1.2:
> >>> > https://github.com/apache/incubator-mxnet/pull/10311
> >>> >
> >>> > Anirudh
> >>> >
> >>> > On Wed, Jun 13, 2018 at 10:00 PM, Marco de Abreu <
> >>> > marco.g.ab...@googlemail.com.invalid> wrote:
> >>> >
> >>> > > I can confirm that this segfaulting test has a big impact.
> >>> > >
> >>> > > On Wed, Jun 13, 2018 at 9:39 PM Aaron Markham
> >>> > >  >>> > >
> >>> > > wrote:
> >>> > >
> >>> > > > I'd keep an eye on this one...  Flaky test:
> test_gru_bidirectional
> >>> > #11219
> >>> > > >
> >>> > > > https://github.com/apache/incubator-mxnet/issues/11219
> >>> > > >
> >>> > > > Just reran several PR's CI runs that all had the same error!
> >>> > > >
> >>> > > > On Wed, Jun 13, 2018 at 5:42 PM, Anirudh 
> >>> > wrote:
> >>> > > >
> >>> > > > > Hi,
> >>> > > > >
> >>> > > > > PRs still in progress : #11127, #11236, #11210, #11054, 
#11216.
> >>> > > > >
> >>> > > > > We are currently facing two issues which are delaying the 
merge
> >>> > > > > of
> >>> > some
> >>> > > > of
> >>> > > > > these PRs:
> >>> > > > > 1. Flaky tests for scala API. A PR is already out to disable
> the
> >>> > test:
> >>> > > > > https://github.com/apache/incubator-mxnet/issues/11249
> >>> > > > > 2. Builds breaking on windows:
> >>> > > > > https://github.com/apache/incubator-mxnet/issues/11265
> >>> > > > >
> >>> > > > > Anirudh
> >>> > > > >
> >>> > > > >
> >>> > > > > On Tue, Jun 12, 2018 at 11:59 AM, Anirudh
> >>> > > > > 
> >>> > > wrote:
> >>> > > > >
> >>> > > > > > Hi all,
> >>> > > > > >
> >>> > > > > > Here are the PRs that are being tracked for 1.2.1 release:
> >>> > > > > >
> >>> > > > > > Related to the save_params backwards incompatible change:
> >>> > > > > > #11127
> >>> > (In
> >>> > > > > > Progress), #11236 (In Progress), #11210 (In Progress) MKLDNN
> >>> > > > > > Fixes: #11212 (In Progress) Cross compilation for armv7 :
> >>> > > > > > #11054 (In Progress) Scala Inference Memory leak fix: #11216
> >>> > > > > > (In Progress) Docs changes: #11211 (Merged) Inplace RELU
> >>> > > > > > Activation, Slice operator perf improvement: #11142
> >>> > > > (Merged)
> >>> > > > > > Use cudnnv7 for depthwise conv #11233 (Merged)
> >>> > > > > >
> >>> > > > > > Please let me know if I have missed something.
> >>> > > > > >
> >>> > > > > > Anirudh
> >>> > > > > >
> >>> > > > 

Re: Clojure Package

[Hagay Lupesko]

Carin,

Thanks for contributing so much to MXNet!
I just went through the planned MXNet v1.3 release scope [1], and did not
see the Clojure package. I think it would be a great addition to MXNet, and
initially can be marked as experimental.

Did you consider adding it to MXNet 1.3 as an experimental feature?

Hagay

[1]
https://cwiki.apache.org/confluence/display/MXNET/Project+Proposals+for+next+MXNet+Release


On Fri, Jun 15, 2018 at 5:52 AM Carin Meier  wrote:

> Kovas Boguta https://twitter.com/kovasb, from the Clojure/AI community,
> graciously took some time to review the PR for the clojure package.
>
> He had some insightful feedback and high level questions that I thought
> might be of interest to the larger dev mailing list.
>
> https://github.com/apache/incubator-mxnet/pull/11205
>
> Feel free to join in on the discussion here or on the PR.
>
> - Carin
>
> On Mon, Jun 11, 2018 at 6:49 PM, Carin Meier  wrote:
>
> > I'm fine with whatever process works best and what makes everyone the
> most
> > comfortable.
> >
> > I've completed one of the requests for code coverage and integrated the
> > cloverage, (https://github.com/cloverage/cloverage), plugin in the PR
> > branch. I've published the results to the confluence page to help with
> > transparency and give everyone a better idea where the level of testing
> is
> > currently.
> > https://cwiki.apache.org/confluence/display/MXNET/
> > Clojure+Package+Contribution+Needs
> >
> >
> >
> > On Mon, Jun 11, 2018 at 6:24 PM, Marco de Abreu <
> > marco.g.ab...@googlemail.com> wrote:
> >
> >> Exactly, that's what I'm proposing. Having the migration in multiple
> >> separate PRs which are done in sequence. This would mean that the
> initial
> >> PRs might not be tested.
> >> We could make the factors you mentioned acceptance criteria to be moved
> >> out
> >> of contrib.
> >>
> >> -Marco
> >>
> >> On Mon, Jun 11, 2018 at 3:17 PM Naveen Swamy 
> wrote:
> >>
> >> > I did not suggest to do it at once. I am not comfortable to merge code
> >> > without good tests, documentation and examples that is not to say
> >> Clojure
> >> > codebase does not have that.
> >> > All that you are saying can happen in separate PRs if you want to
> break
> >> it
> >> > up.
> >> >
> >> >
> >> > On Tue, Jun 12, 2018 at 12:07 AM, Marco de Abreu <
> >> > marco.g.ab...@googlemail.com> wrote:
> >> >
> >> > > The problem I see here is that the migration will have different
> type
> >> of
> >> > > challenges which should be handled isolated. Trying to solve them
> all
> >> at
> >> > > once will make it very lengthy and also hard to review. Considering
> >> Carin
> >> > > and her team this is doing this on a voluntary base, I'd like to
> keep
> >> the
> >> > > number of hoops to jump through per stage as small as possible and
> >> rather
> >> > > split it up into multiple efforts.
> >> > >
> >> > > If we would do everything at once, we'd have to involve a lot of
> >> people
> >> > and
> >> > > it would be hard to review. We'd need at least two or three
> reviewers
> >> > > involved in that process: You (or another committer familiar with
> >> Scala
> >> > to
> >> > > review the Scala part), Yi Zhi (general reviewer) and me (CI
> >> > integration).
> >> > > It would probably require even more committers for other stuff that
> >> comes
> >> > > up. It would rather be better to keep the parts that have to be
> >> touched
> >> > as
> >> > > isolated and few as possible.
> >> > > For example, after the code has been approved and merged in
> general, I
> >> > can
> >> > > assist with the CI integration. This would not require oversight
> from
> >> > other
> >> > > committers, so they'd be free. After that, we'd need somebody
> familiar
> >> > with
> >> > > the release process (probably Naveen) and I'd be free after that.
> >> Then we
> >> > > need general improvements which would also involve other people
> again.
> >> > > Trying to squeeze everything into a single stage is going to make it
> >> very
> >> > > hard in my opinion.
> >> > >
> >> > >
> >> > > -Marco
> >> > >
> >> > > On Mon, Jun 11, 2018 at 2:56 PM Naveen Swamy 
> >> wrote:
> >> > >
> >> > > > I disagree with your approach, We should bring features
> iteratively
> >> > well
> >> > > > tested and well documented. MXNet already has many different
> >> language
> >> > > > bindings which has quite a bit of tech-debt, I don't want just to
> >> add
> >> > > more
> >> > > > tech-debt to the code base with new language bindings as well.
> >> > > >
> >> > > > On Mon, Jun 11, 2018 at 11:17 PM, Marco de Abreu <
> >> > > > marco.g.ab...@googlemail.com> wrote:
> >> > > >
> >> > > > > I think we should try to separate this migration into multiple
> >> > stages:
> >> > > > > 1. Move into contrib
> >> > > > > 2. Migrate release pipeline
> >> > > > > 3. Migrate tests
> >> > > > > 4. Improvements
> >> > > > > 5. Mark as stable and announce Julia officially
> >> > > > >
> >> > > > > I know how much effort Carin and the other maintainers are
> putting
> 

Re: users@mxnet

[Mu Li]

I like the idea to have a user mail list, it should be able to cover the
user segment who prefer to use mail lists. Though my concern is how many of
us will be volunteered to maintain this list. We have several contributors
actively maintain the forum, including Sina, Thom, and Thomas. We need more
volunteers if adding another mail list to guarantee great user experience.



On Mon, Jun 18, 2018 at 11:53 AM, Eric Xie  wrote:

> Neither TF nor Pytorch uses mailing lists though. In fact I can't think of
> any deep learning community that uses mailing lists. Mailing list is an
> obsolete legacy for old projects. No point in bringing it into new projects.
>
> Thanks,
> Eric
>
> On 2018/06/18 18:42:12, Sebastian  wrote:
> > I am puzzled by the reluctance of this project to setup a user
> > mailinglist to be honest.
> >
> > MXNet has major issues with attracting a community outside of Amazon
> > (whenever I hear folks talking about deep learning, they usually mention
> > tensorflow, pytorch and keras, but I rarely hear someone talk about
> > MXNet). At the same time, there is so much resistance to adopt practices
> > that are successfully used by many high-profile toplevel projects...
> >
> > -s
> >
> > On 18.06.2018 20:37, Timur Shenkao wrote:
> > > Facebook is definitely a bad idea: we will be dependent on third party
> > > provider + unclear who & how manages such group etc.
> > > Forum + Confluence + Slack is  much better then.
> > >
> > > On Mon, Jun 18, 2018 at 7:17 PM, Ivan Serdyuk <
> local.tourist.k...@gmail.com>
> > > wrote:
> > >
> > >> Greetings Barber, Christopher. I had an idea to move out some
> discussions,
> > >> covering Java and Scala API, to Facebook. So if somewhere exists a
> local
> > >> JUG or Scala user group - they could reflect the topic of discussion.
> But
> > >> background stuff could take place on mailing lists, Slack, forum,
> whatever.
> > >> The reverse mechanism could be used to involve new committers, as
> well (so
> > >> they would appear as presented newcomers, as for contributions).
> > >>
> > >> Regards
> > >> Ivan
> > >>
> > >> On Mon, Jun 18, 2018 at 8:40 PM, Barber, Christopher <
> > >> christopher.bar...@analog.com> wrote:
> > >>
> > >>> I don't understand why you would want a users mailing list when you
> > >>> already have discussion forums. Users that want to be notified of new
> > >> posts
> > >>> on the forum can configure their notification preferences
> appropriately.
> > >>> The traffic on the forums is already pretty low. I would think you
> would
> > >>> not want to dilute that further.
> > >>>
> > >>> Christopher
> > >>>
> > >>> On 6/18/18, 1:27 PM, "Jim Jagielski"  wrote:
> > >>>
> > >>>  users@ mailing lists have great societal advantages that one
> > >>> shouldn't ignore...
> > >>>
> > >>>  And it's not like this is the only project with "multiple"
> > >>> communication choices for users. Most, if not all, projects have
> users@in
> > >>> addition to such supplemental methods as IRC channels, a forum,
> etc...
> > >> It's
> > >>> about making it easy to have as many users as possible and as many
> > >>> potential ways for users to communicate. It's not confusing; it's
> > >>> empowering :)
> > >>>
> > >>>  > On Jun 18, 2018, at 1:19 PM, Tianqi Chen <
> tqc...@cs.washington.edu
> > >>>
> > >>> wrote:
> > >>>  >
> > >>>  > The problem of having multiple separate channels of
> communication
> > >> is
> > >>> that
> > >>>  > users get confused, and the cost of maintenance goes
> up(people have
> > >>> to
> > >>>  > watch both). As the current community was at discuss forum
> and many
> > >>> users
> > >>>  > prefer it, having a mail-list is only a burden we will bring
> > >>>  >
> > >>>  > Tianqi
> > >>>  >
> > >>>  > On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski <
> j...@jagunet.com>
> > >>> wrote:
> > >>>  >
> > >>>  >> IMO, that is the wrong way to look at it.
> > >>>  >>
> > >>>  >> A users@ mailing list is a great, easy, low-cost and
> low-overhead
> > >>> way of
> > >>>  >> *increasing* the user community and providing an extra level
> of
> > >>> support.
> > >>>  >> Unless there is "strong evidence" that this is NOT the case,
> I
> > >> would
> > >>>  >> recommend we create the list.
> > >>>  >>
> > >>>  >>> On Jun 16, 2018, at 12:28 AM, Tianqi Chen <
> > >>> tqc...@cs.washington.edu>
> > >>>  >> wrote:
> > >>>  >>>
> > >>>  >>> So unless there is a strong evidence that our community
> users
> > >>> prefers the
> > >>>  >>> mail-list, I would recommend we keep the current way
> > >>>  >>>
> > >>>  >>> Tianqi
> > >>>  >>>
> > >>>  >>> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández <
> > >>> wik...@apache.org>
> > >>>  >> wrote:
> > >>>  >>>
> > >>>   Are we targeting just Seattle as our community? I really
> hope we
> > >>> are
> > >>>   thinking a bit beyond that...
> > >>>  
> > >>>   On Fri, 

Re: users@mxnet

[Ivan Serdyuk]

Well, there is a trend to for speaking about MXNet (Ukraine). At least I
have visited a meetup (with a workshop).

On Mon, Jun 18, 2018 at 9:42 PM, Sebastian  wrote:

> I am puzzled by the reluctance of this project to setup a user mailinglist
> to be honest.
>
> MXNet has major issues with attracting a community outside of Amazon
> (whenever I hear folks talking about deep learning, they usually mention
> tensorflow, pytorch and keras, but I rarely hear someone talk about MXNet).
> At the same time, there is so much resistance to adopt practices that are
> successfully used by many high-profile toplevel projects...
>
> -s
>
>
> On 18.06.2018 20:37, Timur Shenkao wrote:
>
>> Facebook is definitely a bad idea: we will be dependent on third party
>> provider + unclear who & how manages such group etc.
>> Forum + Confluence + Slack is  much better then.
>>
>> On Mon, Jun 18, 2018 at 7:17 PM, Ivan Serdyuk <
>> local.tourist.k...@gmail.com>
>> wrote:
>>
>> Greetings Barber, Christopher. I had an idea to move out some discussions,
>>> covering Java and Scala API, to Facebook. So if somewhere exists a local
>>> JUG or Scala user group - they could reflect the topic of discussion. But
>>> background stuff could take place on mailing lists, Slack, forum,
>>> whatever.
>>> The reverse mechanism could be used to involve new committers, as well
>>> (so
>>> they would appear as presented newcomers, as for contributions).
>>>
>>> Regards
>>> Ivan
>>>
>>> On Mon, Jun 18, 2018 at 8:40 PM, Barber, Christopher <
>>> christopher.bar...@analog.com> wrote:
>>>
>>> I don't understand why you would want a users mailing list when you
 already have discussion forums. Users that want to be notified of new

>>> posts
>>>
 on the forum can configure their notification preferences appropriately.
 The traffic on the forums is already pretty low. I would think you would
 not want to dilute that further.

 Christopher

 On 6/18/18, 1:27 PM, "Jim Jagielski"  wrote:

  users@ mailing lists have great societal advantages that one
 shouldn't ignore...

  And it's not like this is the only project with "multiple"
 communication choices for users. Most, if not all, projects have
 users@in
 addition to such supplemental methods as IRC channels, a forum, etc...

>>> It's
>>>
 about making it easy to have as many users as possible and as many
 potential ways for users to communicate. It's not confusing; it's
 empowering :)

  > On Jun 18, 2018, at 1:19 PM, Tianqi Chen <
 tqc...@cs.washington.edu

 wrote:
  >
  > The problem of having multiple separate channels of communication

>>> is
>>>
 that
  > users get confused, and the cost of maintenance goes up(people
 have
 to
  > watch both). As the current community was at discuss forum and
 many
 users
  > prefer it, having a mail-list is only a burden we will bring
  >
  > Tianqi
  >
  > On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski 
 wrote:
  >
  >> IMO, that is the wrong way to look at it.
  >>
  >> A users@ mailing list is a great, easy, low-cost and
 low-overhead
 way of
  >> *increasing* the user community and providing an extra level of
 support.
  >> Unless there is "strong evidence" that this is NOT the case, I

>>> would
>>>
  >> recommend we create the list.
  >>
  >>> On Jun 16, 2018, at 12:28 AM, Tianqi Chen <
 tqc...@cs.washington.edu>
  >> wrote:
  >>>
  >>> So unless there is a strong evidence that our community users
 prefers the
  >>> mail-list, I would recommend we keep the current way
  >>>
  >>> Tianqi
  >>>
  >>> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández <
 wik...@apache.org>
  >> wrote:
  >>>
   Are we targeting just Seattle as our community? I really hope
 we
 are
   thinking a bit beyond that...
  
   On Fri, Jun 15, 2018, 21:22 Tianqi Chen <

>>> tqc...@cs.washington.edu
>>>

>  >> wrote:
  
  > I remember last time during the mxnet meetup in Seattle, we

>>> did a
>>>
  >> survey,
  > and most users preferred the current discuss forum. So I
 would
 say we
   stick
  > with that given the user community prefers that
  >
  > Tianqi
  >
  > On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández <
 wik...@apache.org>
  > wrote:
  >
  >> Then, if everybody agree, let's request the mailing list
 creation to
  > INFRA
  >> ;-)
  >>
  >> Marco, I wouldn't do that. Typically developers are also
 sub

Re: users@mxnet

[Eric Xie]

Neither TF nor Pytorch uses mailing lists though. In fact I can't think of any 
deep learning community that uses mailing lists. Mailing list is an obsolete 
legacy for old projects. No point in bringing it into new projects.

Thanks,
Eric

On 2018/06/18 18:42:12, Sebastian  wrote: 
> I am puzzled by the reluctance of this project to setup a user 
> mailinglist to be honest.
> 
> MXNet has major issues with attracting a community outside of Amazon 
> (whenever I hear folks talking about deep learning, they usually mention 
> tensorflow, pytorch and keras, but I rarely hear someone talk about 
> MXNet). At the same time, there is so much resistance to adopt practices 
> that are successfully used by many high-profile toplevel projects...
> 
> -s
> 
> On 18.06.2018 20:37, Timur Shenkao wrote:
> > Facebook is definitely a bad idea: we will be dependent on third party
> > provider + unclear who & how manages such group etc.
> > Forum + Confluence + Slack is  much better then.
> > 
> > On Mon, Jun 18, 2018 at 7:17 PM, Ivan Serdyuk 
> > wrote:
> > 
> >> Greetings Barber, Christopher. I had an idea to move out some discussions,
> >> covering Java and Scala API, to Facebook. So if somewhere exists a local
> >> JUG or Scala user group - they could reflect the topic of discussion. But
> >> background stuff could take place on mailing lists, Slack, forum, whatever.
> >> The reverse mechanism could be used to involve new committers, as well (so
> >> they would appear as presented newcomers, as for contributions).
> >>
> >> Regards
> >> Ivan
> >>
> >> On Mon, Jun 18, 2018 at 8:40 PM, Barber, Christopher <
> >> christopher.bar...@analog.com> wrote:
> >>
> >>> I don't understand why you would want a users mailing list when you
> >>> already have discussion forums. Users that want to be notified of new
> >> posts
> >>> on the forum can configure their notification preferences appropriately.
> >>> The traffic on the forums is already pretty low. I would think you would
> >>> not want to dilute that further.
> >>>
> >>> Christopher
> >>>
> >>> On 6/18/18, 1:27 PM, "Jim Jagielski"  wrote:
> >>>
> >>>  users@ mailing lists have great societal advantages that one
> >>> shouldn't ignore...
> >>>
> >>>  And it's not like this is the only project with "multiple"
> >>> communication choices for users. Most, if not all, projects have users@in
> >>> addition to such supplemental methods as IRC channels, a forum, etc...
> >> It's
> >>> about making it easy to have as many users as possible and as many
> >>> potential ways for users to communicate. It's not confusing; it's
> >>> empowering :)
> >>>
> >>>  > On Jun 18, 2018, at 1:19 PM, Tianqi Chen  >>>
> >>> wrote:
> >>>  >
> >>>  > The problem of having multiple separate channels of communication
> >> is
> >>> that
> >>>  > users get confused, and the cost of maintenance goes up(people have
> >>> to
> >>>  > watch both). As the current community was at discuss forum and many
> >>> users
> >>>  > prefer it, having a mail-list is only a burden we will bring
> >>>  >
> >>>  > Tianqi
> >>>  >
> >>>  > On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski 
> >>> wrote:
> >>>  >
> >>>  >> IMO, that is the wrong way to look at it.
> >>>  >>
> >>>  >> A users@ mailing list is a great, easy, low-cost and low-overhead
> >>> way of
> >>>  >> *increasing* the user community and providing an extra level of
> >>> support.
> >>>  >> Unless there is "strong evidence" that this is NOT the case, I
> >> would
> >>>  >> recommend we create the list.
> >>>  >>
> >>>  >>> On Jun 16, 2018, at 12:28 AM, Tianqi Chen <
> >>> tqc...@cs.washington.edu>
> >>>  >> wrote:
> >>>  >>>
> >>>  >>> So unless there is a strong evidence that our community users
> >>> prefers the
> >>>  >>> mail-list, I would recommend we keep the current way
> >>>  >>>
> >>>  >>> Tianqi
> >>>  >>>
> >>>  >>> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández <
> >>> wik...@apache.org>
> >>>  >> wrote:
> >>>  >>>
> >>>   Are we targeting just Seattle as our community? I really hope we
> >>> are
> >>>   thinking a bit beyond that...
> >>>  
> >>>   On Fri, Jun 15, 2018, 21:22 Tianqi Chen <
> >> tqc...@cs.washington.edu
> 
> >>>  >> wrote:
> >>>  
> >>>  > I remember last time during the mxnet meetup in Seattle, we
> >> did a
> >>>  >> survey,
> >>>  > and most users preferred the current discuss forum. So I would
> >>> say we
> >>>   stick
> >>>  > with that given the user community prefers that
> >>>  >
> >>>  > Tianqi
> >>>  >
> >>>  > On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández <
> >>> wik...@apache.org>
> >>>  > wrote:
> >>>  >
> >>>  >> Then, if everybody agree, let's request the mailing list
> >>> creation to
> >>>  > INFRA
> >>>  >> ;-)
> >>>  >>
> >>

Re: users@mxnet

[Tianqi Chen]

IMHO, the approach(mail list or discuss) have nothing to do with the
popularity of the project.  If you look at TF or pytorch you mentioned.
Pytorch uses discuss forum and slack, tf uses stackoverflow for support.
Both are popular but not adopting maillist. Note that I know both are both
not apache projects, but just to show that the popularity of the project do
not necessarily have to go with the setup of maillist


Tianqi


On Mon, Jun 18, 2018 at 11:42 AM Sebastian  wrote:

> I am puzzled by the reluctance of this project to setup a user
> mailinglist to be honest.
>
> MXNet has major issues with attracting a community outside of Amazon
> (whenever I hear folks talking about deep learning, they usually mention
> tensorflow, pytorch and keras, but I rarely hear someone talk about
> MXNet). At the same time, there is so much resistance to adopt practices
> that are successfully used by many high-profile toplevel projects...
>
> -s
>
> On 18.06.2018 20:37, Timur Shenkao wrote:
> > Facebook is definitely a bad idea: we will be dependent on third party
> > provider + unclear who & how manages such group etc.
> > Forum + Confluence + Slack is  much better then.
> >
> > On Mon, Jun 18, 2018 at 7:17 PM, Ivan Serdyuk <
> local.tourist.k...@gmail.com>
> > wrote:
> >
> >> Greetings Barber, Christopher. I had an idea to move out some
> discussions,
> >> covering Java and Scala API, to Facebook. So if somewhere exists a local
> >> JUG or Scala user group - they could reflect the topic of discussion.
> But
> >> background stuff could take place on mailing lists, Slack, forum,
> whatever.
> >> The reverse mechanism could be used to involve new committers, as well
> (so
> >> they would appear as presented newcomers, as for contributions).
> >>
> >> Regards
> >> Ivan
> >>
> >> On Mon, Jun 18, 2018 at 8:40 PM, Barber, Christopher <
> >> christopher.bar...@analog.com> wrote:
> >>
> >>> I don't understand why you would want a users mailing list when you
> >>> already have discussion forums. Users that want to be notified of new
> >> posts
> >>> on the forum can configure their notification preferences
> appropriately.
> >>> The traffic on the forums is already pretty low. I would think you
> would
> >>> not want to dilute that further.
> >>>
> >>> Christopher
> >>>
> >>> On 6/18/18, 1:27 PM, "Jim Jagielski"  wrote:
> >>>
> >>>  users@ mailing lists have great societal advantages that one
> >>> shouldn't ignore...
> >>>
> >>>  And it's not like this is the only project with "multiple"
> >>> communication choices for users. Most, if not all, projects have
> users@in
> >>> addition to such supplemental methods as IRC channels, a forum, etc...
> >> It's
> >>> about making it easy to have as many users as possible and as many
> >>> potential ways for users to communicate. It's not confusing; it's
> >>> empowering :)
> >>>
> >>>  > On Jun 18, 2018, at 1:19 PM, Tianqi Chen <
> tqc...@cs.washington.edu
> >>>
> >>> wrote:
> >>>  >
> >>>  > The problem of having multiple separate channels of
> communication
> >> is
> >>> that
> >>>  > users get confused, and the cost of maintenance goes up(people
> have
> >>> to
> >>>  > watch both). As the current community was at discuss forum and
> many
> >>> users
> >>>  > prefer it, having a mail-list is only a burden we will bring
> >>>  >
> >>>  > Tianqi
> >>>  >
> >>>  > On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski  >
> >>> wrote:
> >>>  >
> >>>  >> IMO, that is the wrong way to look at it.
> >>>  >>
> >>>  >> A users@ mailing list is a great, easy, low-cost and
> low-overhead
> >>> way of
> >>>  >> *increasing* the user community and providing an extra level of
> >>> support.
> >>>  >> Unless there is "strong evidence" that this is NOT the case, I
> >> would
> >>>  >> recommend we create the list.
> >>>  >>
> >>>  >>> On Jun 16, 2018, at 12:28 AM, Tianqi Chen <
> >>> tqc...@cs.washington.edu>
> >>>  >> wrote:
> >>>  >>>
> >>>  >>> So unless there is a strong evidence that our community users
> >>> prefers the
> >>>  >>> mail-list, I would recommend we keep the current way
> >>>  >>>
> >>>  >>> Tianqi
> >>>  >>>
> >>>  >>> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández <
> >>> wik...@apache.org>
> >>>  >> wrote:
> >>>  >>>
> >>>   Are we targeting just Seattle as our community? I really
> hope we
> >>> are
> >>>   thinking a bit beyond that...
> >>>  
> >>>   On Fri, Jun 15, 2018, 21:22 Tianqi Chen <
> >> tqc...@cs.washington.edu
> 
> >>>  >> wrote:
> >>>  
> >>>  > I remember last time during the mxnet meetup in Seattle, we
> >> did a
> >>>  >> survey,
> >>>  > and most users preferred the current discuss forum. So I
> would
> >>> say we
> >>>   stick
> >>>  > with that given the user community prefers that
> >>>  >
> >>>  > Tianqi
> >>>  >
> >>>  >

Re: users@mxnet

[Barber, Christopher]

Whatever you do, make sure to list all these information sources in one 
easy-to-find place. For instance, it may not be very obvious to anyone that 
they can read the dev mailing list on lists.apache.org. It is bad if users 
aren't even aware that other channels exist.

On 6/18/18, 2:45 PM, "Yasser Zamani"  wrote:



On 6/18/2018 9:18 PM, Jim Jagielski wrote:
> IMO, that is the wrong way to look at it.
> 
> A users@ mailing list is a great, easy, low-cost and low-overhead way of 
*increasing* the user community and providing an extra level of support. Unless 
there is "strong evidence" that this is NOT the case, I would recommend we 
create the list.


As an already Apache Committer of Struts, I also prefer mail list like Jim.

I fell in love with MXNet and I'm still learning hard to being able to
have contributions one day. It was somehow hard for me to also register
and daily monitor another system (your user forum). My feel was same as
your current feel about a new mail list, but, I remembered one rule in
Apache Way: "At Apache, deciders are who do the work". And as I didn't
have any contribution till that time, then I thought I must respect your
decision, so I registered there!

However, I personally still recommend sticking together with Apache
INFRA under Apache Way umbrella as much as possible. I think these help
us to standardize things, understanding each other better and finally
get a graduation from Apache Incubator. Anyway, finally, you who have
done or do the work, are deciders :)

Best Regards.

Re: users@mxnet

[Yasser Zamani]

On 6/18/2018 9:18 PM, Jim Jagielski wrote:
> IMO, that is the wrong way to look at it.
> 
> A users@ mailing list is a great, easy, low-cost and low-overhead way of 
> *increasing* the user community and providing an extra level of support. 
> Unless there is "strong evidence" that this is NOT the case, I would 
> recommend we create the list.


As an already Apache Committer of Struts, I also prefer mail list like Jim.

I fell in love with MXNet and I'm still learning hard to being able to
have contributions one day. It was somehow hard for me to also register
and daily monitor another system (your user forum). My feel was same as
your current feel about a new mail list, but, I remembered one rule in
Apache Way: "At Apache, deciders are who do the work". And as I didn't
have any contribution till that time, then I thought I must respect your
decision, so I registered there!

However, I personally still recommend sticking together with Apache
INFRA under Apache Way umbrella as much as possible. I think these help
us to standardize things, understanding each other better and finally
get a graduation from Apache Incubator. Anyway, finally, you who have
done or do the work, are deciders :)

Best Regards.

Re: users@mxnet

[Sebastian]

I am puzzled by the reluctance of this project to setup a user 
mailinglist to be honest.


MXNet has major issues with attracting a community outside of Amazon 
(whenever I hear folks talking about deep learning, they usually mention 
tensorflow, pytorch and keras, but I rarely hear someone talk about 
MXNet). At the same time, there is so much resistance to adopt practices 
that are successfully used by many high-profile toplevel projects...


-s

On 18.06.2018 20:37, Timur Shenkao wrote:

Facebook is definitely a bad idea: we will be dependent on third party
provider + unclear who & how manages such group etc.
Forum + Confluence + Slack is  much better then.

On Mon, Jun 18, 2018 at 7:17 PM, Ivan Serdyuk 
wrote:


Greetings Barber, Christopher. I had an idea to move out some discussions,
covering Java and Scala API, to Facebook. So if somewhere exists a local
JUG or Scala user group - they could reflect the topic of discussion. But
background stuff could take place on mailing lists, Slack, forum, whatever.
The reverse mechanism could be used to involve new committers, as well (so
they would appear as presented newcomers, as for contributions).

Regards
Ivan

On Mon, Jun 18, 2018 at 8:40 PM, Barber, Christopher <
christopher.bar...@analog.com> wrote:


I don't understand why you would want a users mailing list when you
already have discussion forums. Users that want to be notified of new

posts

on the forum can configure their notification preferences appropriately.
The traffic on the forums is already pretty low. I would think you would
not want to dilute that further.

Christopher

On 6/18/18, 1:27 PM, "Jim Jagielski"  wrote:

 users@ mailing lists have great societal advantages that one
shouldn't ignore...

 And it's not like this is the only project with "multiple"
communication choices for users. Most, if not all, projects have users@in
addition to such supplemental methods as IRC channels, a forum, etc...

It's

about making it easy to have as many users as possible and as many
potential ways for users to communicate. It's not confusing; it's
empowering :)

 > On Jun 18, 2018, at 1:19 PM, Tianqi Chen 
 > The problem of having multiple separate channels of communication

is

that
 > users get confused, and the cost of maintenance goes up(people have
to
 > watch both). As the current community was at discuss forum and many
users
 > prefer it, having a mail-list is only a burden we will bring
 >
 > Tianqi
 >
 > On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski 
wrote:
 >
 >> IMO, that is the wrong way to look at it.
 >>
 >> A users@ mailing list is a great, easy, low-cost and low-overhead
way of
 >> *increasing* the user community and providing an extra level of
support.
 >> Unless there is "strong evidence" that this is NOT the case, I

would

 >> recommend we create the list.
 >>
 >>> On Jun 16, 2018, at 12:28 AM, Tianqi Chen <
tqc...@cs.washington.edu>
 >> wrote:
 >>>
 >>> So unless there is a strong evidence that our community users
prefers the
 >>> mail-list, I would recommend we keep the current way
 >>>
 >>> Tianqi
 >>>
 >>> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández <
wik...@apache.org>
 >> wrote:
 >>>
  Are we targeting just Seattle as our community? I really hope we
are
  thinking a bit beyond that...
 
  On Fri, Jun 15, 2018, 21:22 Tianqi Chen <

tqc...@cs.washington.edu



 >> wrote:
 
 > I remember last time during the mxnet meetup in Seattle, we

did a

 >> survey,
 > and most users preferred the current discuss forum. So I would
say we
  stick
 > with that given the user community prefers that
 >
 > Tianqi
 >
 > On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández <
wik...@apache.org>
 > wrote:
 >
 >> Then, if everybody agree, let's request the mailing list
creation to
 > INFRA
 >> ;-)
 >>
 >> Marco, I wouldn't do that. Typically developers are also
subscribed
 > there,
 >> since they may be the most informed people for answering

users'

 > questions.
 >> But the topics discussed there may not be of the interest for
pure
 >> development purposes. Some discussions will jump from users@
to dev@,
 > but
 >> at a different level. So I wouldn't forward one mailing list

to

the
 > other.
 >>
 >> On Fri, Jun 15, 2018, 21:01 Marco de Abreu
 >>  wrote:
 >>
 >>> I think nobody was opposed to it in the past, right?
 >>>
 >>> I'd propose that all emails automatically get copied to dev@
to
  ensure
 >>> high
 >>> visibility initially. What do you think?
 >>>
 >>> Sebastian  schrieb am Fr., 15. Juni 2018,
20:51:
 >>>
  I have already pr

Re: Config for Scala Maven Package

[Anirudh]

Hi Naveen,

Sorry I missed this. Can you please elaborate on how long it takes
typically for publishing the scala packages to maven before the vote?
I think if there is considerable churn in publishing these packages today,
we should not wait for the package for the vote as we may sacrifice our big
python user base
getting the release early for the scala user base.

Anirudh


On Thu, May 31, 2018 at 7:27 AM, Naveen Swamy  wrote:

> All, I have created a guide to maven publish process here;
> https://cwiki.apache.org/confluence/display/MXNET/
> MXNet-Scala+Release+Process
> .
> Going forward, I propose that we include the scala maven packages also in
> our RC voting(I have updated the release-process already), I understand the
> maven packages use the same code base that is sent for voting. I
> encountered a quite bit of issues trying to publish the package, this could
> because we never did this past and I have automated as much as possible. I
> think there is value in getting these packages tested during the RC
> process.
>
> Sandeep,
> I am using what is available on AWS Deep Learning Base AMI which is 3.4 and
> CUDNN 7.0.5.
>
> Thanks, Naveen
>
> On Wed, May 23, 2018 at 8:57 AM, sandeep krishnamurthy <
> sandeep.krishn...@gmail.com> wrote:
>
> > Hi Naveen,
> >
> > Great work.
> > I was wondering about OpenCV and cuDNN
> >
> > Thanks,
> > Sandeep
> >
> > On Mon, May 21, 2018 at 5:44 PM, Naveen Swamy 
> wrote:
> >
> > > Hi Marco,
> > > For MKL/MKLDNN --> I don't think we should build since its still
> > > experimental. Its sad that MKLML was removed without a proper
> discussion,
> > > granted
> > > I have not tested with CUDA 9.2, I will building on a AWS Ubuntu
> instance
> > > that is built with CUDA 9.0, I presume it should work fine.
> > >
> > >
> > > On Mon, May 21, 2018 at 5:31 PM, Marco de Abreu <
> > > marco.g.ab...@googlemail.com> wrote:
> > >
> > > > Hi Naveen,
> > > >
> > > > thank you for driving the releases for the Scala packages!
> > > >
> > > > For CPU, I'd propose add MKL/MKLDNN builds.
> > > >
> > > > For GPU, do we have to specify the minor version of CUDA up front or
> > are
> > > > they interchangeable? CUDA 9.2, for example, received some quite
> > > > significant performance updates and it would be great if our users
> > could
> > > > use them.
> > > >
> > > > Best regards,
> > > > Marco
> > > >
> > > > On Fri, May 18, 2018 at 8:28 PM, Naveen Swamy 
> > > wrote:
> > > >
> > > > > I am working on publishing the Scala package from 1.2.0 release.
> What
> > > do
> > > > > you recommend should be the configuration for the Scala Packages ?
> > > > >
> > > > > I am thinking of the below
> > > > > For CPU build with
> > > > > OPENBLAS
> > > > > LAPACK
> > > > >
> > > > > For GPU build with
> > > > > OPENBLAS
> > > > > CUDA 9.0
> > > > > NCCL
> > > > >
> > > > > Let me know if you concerns or recommendations.
> > > > >
> > > > > Thanks, Naveen
> > > > >
> > > >
> > >
> >
> >
> >
> > --
> > Sandeep Krishnamurthy
> >
>

Re: users@mxnet

[Timur Shenkao]

Facebook is definitely a bad idea: we will be dependent on third party
provider + unclear who & how manages such group etc.
Forum + Confluence + Slack is  much better then.

On Mon, Jun 18, 2018 at 7:17 PM, Ivan Serdyuk 
wrote:

> Greetings Barber, Christopher. I had an idea to move out some discussions,
> covering Java and Scala API, to Facebook. So if somewhere exists a local
> JUG or Scala user group - they could reflect the topic of discussion. But
> background stuff could take place on mailing lists, Slack, forum, whatever.
> The reverse mechanism could be used to involve new committers, as well (so
> they would appear as presented newcomers, as for contributions).
>
> Regards
> Ivan
>
> On Mon, Jun 18, 2018 at 8:40 PM, Barber, Christopher <
> christopher.bar...@analog.com> wrote:
>
> > I don't understand why you would want a users mailing list when you
> > already have discussion forums. Users that want to be notified of new
> posts
> > on the forum can configure their notification preferences appropriately.
> > The traffic on the forums is already pretty low. I would think you would
> > not want to dilute that further.
> >
> > Christopher
> >
> > On 6/18/18, 1:27 PM, "Jim Jagielski"  wrote:
> >
> > users@ mailing lists have great societal advantages that one
> > shouldn't ignore...
> >
> > And it's not like this is the only project with "multiple"
> > communication choices for users. Most, if not all, projects have users@in
> > addition to such supplemental methods as IRC channels, a forum, etc...
> It's
> > about making it easy to have as many users as possible and as many
> > potential ways for users to communicate. It's not confusing; it's
> > empowering :)
> >
> > > On Jun 18, 2018, at 1:19 PM, Tianqi Chen  >
> > wrote:
> > >
> > > The problem of having multiple separate channels of communication
> is
> > that
> > > users get confused, and the cost of maintenance goes up(people have
> > to
> > > watch both). As the current community was at discuss forum and many
> > users
> > > prefer it, having a mail-list is only a burden we will bring
> > >
> > > Tianqi
> > >
> > > On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski 
> > wrote:
> > >
> > >> IMO, that is the wrong way to look at it.
> > >>
> > >> A users@ mailing list is a great, easy, low-cost and low-overhead
> > way of
> > >> *increasing* the user community and providing an extra level of
> > support.
> > >> Unless there is "strong evidence" that this is NOT the case, I
> would
> > >> recommend we create the list.
> > >>
> > >>> On Jun 16, 2018, at 12:28 AM, Tianqi Chen <
> > tqc...@cs.washington.edu>
> > >> wrote:
> > >>>
> > >>> So unless there is a strong evidence that our community users
> > prefers the
> > >>> mail-list, I would recommend we keep the current way
> > >>>
> > >>> Tianqi
> > >>>
> > >>> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández <
> > wik...@apache.org>
> > >> wrote:
> > >>>
> >  Are we targeting just Seattle as our community? I really hope we
> > are
> >  thinking a bit beyond that...
> > 
> >  On Fri, Jun 15, 2018, 21:22 Tianqi Chen <
> tqc...@cs.washington.edu
> > >
> > >> wrote:
> > 
> > > I remember last time during the mxnet meetup in Seattle, we
> did a
> > >> survey,
> > > and most users preferred the current discuss forum. So I would
> > say we
> >  stick
> > > with that given the user community prefers that
> > >
> > > Tianqi
> > >
> > > On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández <
> > wik...@apache.org>
> > > wrote:
> > >
> > >> Then, if everybody agree, let's request the mailing list
> > creation to
> > > INFRA
> > >> ;-)
> > >>
> > >> Marco, I wouldn't do that. Typically developers are also
> > subscribed
> > > there,
> > >> since they may be the most informed people for answering
> users'
> > > questions.
> > >> But the topics discussed there may not be of the interest for
> > pure
> > >> development purposes. Some discussions will jump from users@
> > to dev@,
> > > but
> > >> at a different level. So I wouldn't forward one mailing list
> to
> > the
> > > other.
> > >>
> > >> On Fri, Jun 15, 2018, 21:01 Marco de Abreu
> > >>  wrote:
> > >>
> > >>> I think nobody was opposed to it in the past, right?
> > >>>
> > >>> I'd propose that all emails automatically get copied to dev@
> > to
> >  ensure
> > >>> high
> > >>> visibility initially. What do you think?
> > >>>
> > >>> Sebastian  schrieb am Fr., 15. Juni 2018,
> > 20:51:
> > >>>
> >  I have already proposed this many times in the past and
> would
> > > strongly
> > >>

Single-Machine Topology-aware Communication

[Carl Yang]

Hi,

Currently, we have two methods for single-machine communication:
parameter server and NCCL ring reduction. Both of these methods have
some downsides. Parameter server does not differentiate between NVLink
connections and PCI-E, so it ends up using the higher latency and
slower PCI-E connections as frequently as it does NVLink. NCCL uses
the ring reduce algorithm, which has higher theoretical latency than
other algorithms. NCCL also requires users to install another
dependency in order to use it. I am working on a topology-aware
approach that can
address these limitations. Design proposal is on cwiki:
https://cwiki.apache.org/confluence/display/MXNET/Single+machine+All+Reduce+Topology-aware+Communication

Please feel free to let me know if you have any suggestions.

Regards,
Carl

Re: users@mxnet

[Ivan Serdyuk]

Greetings Barber, Christopher. I had an idea to move out some discussions,
covering Java and Scala API, to Facebook. So if somewhere exists a local
JUG or Scala user group - they could reflect the topic of discussion. But
background stuff could take place on mailing lists, Slack, forum, whatever.
The reverse mechanism could be used to involve new committers, as well (so
they would appear as presented newcomers, as for contributions).

Regards
Ivan

On Mon, Jun 18, 2018 at 8:40 PM, Barber, Christopher <
christopher.bar...@analog.com> wrote:

> I don't understand why you would want a users mailing list when you
> already have discussion forums. Users that want to be notified of new posts
> on the forum can configure their notification preferences appropriately.
> The traffic on the forums is already pretty low. I would think you would
> not want to dilute that further.
>
> Christopher
>
> On 6/18/18, 1:27 PM, "Jim Jagielski"  wrote:
>
> users@ mailing lists have great societal advantages that one
> shouldn't ignore...
>
> And it's not like this is the only project with "multiple"
> communication choices for users. Most, if not all, projects have users@in
> addition to such supplemental methods as IRC channels, a forum, etc... It's
> about making it easy to have as many users as possible and as many
> potential ways for users to communicate. It's not confusing; it's
> empowering :)
>
> > On Jun 18, 2018, at 1:19 PM, Tianqi Chen 
> wrote:
> >
> > The problem of having multiple separate channels of communication is
> that
> > users get confused, and the cost of maintenance goes up(people have
> to
> > watch both). As the current community was at discuss forum and many
> users
> > prefer it, having a mail-list is only a burden we will bring
> >
> > Tianqi
> >
> > On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski 
> wrote:
> >
> >> IMO, that is the wrong way to look at it.
> >>
> >> A users@ mailing list is a great, easy, low-cost and low-overhead
> way of
> >> *increasing* the user community and providing an extra level of
> support.
> >> Unless there is "strong evidence" that this is NOT the case, I would
> >> recommend we create the list.
> >>
> >>> On Jun 16, 2018, at 12:28 AM, Tianqi Chen <
> tqc...@cs.washington.edu>
> >> wrote:
> >>>
> >>> So unless there is a strong evidence that our community users
> prefers the
> >>> mail-list, I would recommend we keep the current way
> >>>
> >>> Tianqi
> >>>
> >>> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández <
> wik...@apache.org>
> >> wrote:
> >>>
>  Are we targeting just Seattle as our community? I really hope we
> are
>  thinking a bit beyond that...
> 
>  On Fri, Jun 15, 2018, 21:22 Tianqi Chen  >
> >> wrote:
> 
> > I remember last time during the mxnet meetup in Seattle, we did a
> >> survey,
> > and most users preferred the current discuss forum. So I would
> say we
>  stick
> > with that given the user community prefers that
> >
> > Tianqi
> >
> > On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández <
> wik...@apache.org>
> > wrote:
> >
> >> Then, if everybody agree, let's request the mailing list
> creation to
> > INFRA
> >> ;-)
> >>
> >> Marco, I wouldn't do that. Typically developers are also
> subscribed
> > there,
> >> since they may be the most informed people for answering users'
> > questions.
> >> But the topics discussed there may not be of the interest for
> pure
> >> development purposes. Some discussions will jump from users@
> to dev@,
> > but
> >> at a different level. So I wouldn't forward one mailing list to
> the
> > other.
> >>
> >> On Fri, Jun 15, 2018, 21:01 Marco de Abreu
> >>  wrote:
> >>
> >>> I think nobody was opposed to it in the past, right?
> >>>
> >>> I'd propose that all emails automatically get copied to dev@
> to
>  ensure
> >>> high
> >>> visibility initially. What do you think?
> >>>
> >>> Sebastian  schrieb am Fr., 15. Juni 2018,
> 20:51:
> >>>
>  I have already proposed this many times in the past and would
> > strongly
>  encourage it.
> 
>  -s
> 
>  On 15.06.2018 21:56, Sergio Fernández wrote:
> > Hi,
> >
> > is there any good reason why the podling doesn't have a
> users@
> >> mailing
>  list
> > yet?
> >
> > Honestly speaking, I'm not a big fan of the other tools the
>  podling
> >> is
> > using. Slack and Web forums a cool tools, and I used them a
> lot
>  in
>

Re: users@mxnet

[Barber, Christopher]

I don't understand why you would want a users mailing list when you already 
have discussion forums. Users that want to be notified of new posts on the 
forum can configure their notification preferences appropriately. The traffic 
on the forums is already pretty low. I would think you would not want to dilute 
that further.

Christopher 

On 6/18/18, 1:27 PM, "Jim Jagielski"  wrote:

users@ mailing lists have great societal advantages that one shouldn't 
ignore...

And it's not like this is the only project with "multiple" communication 
choices for users. Most, if not all, projects have users@in addition to such 
supplemental methods as IRC channels, a forum, etc... It's about making it easy 
to have as many users as possible and as many potential ways for users to 
communicate. It's not confusing; it's empowering :)

> On Jun 18, 2018, at 1:19 PM, Tianqi Chen  wrote:
> 
> The problem of having multiple separate channels of communication is that
> users get confused, and the cost of maintenance goes up(people have to
> watch both). As the current community was at discuss forum and many users
> prefer it, having a mail-list is only a burden we will bring
> 
> Tianqi
> 
> On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski  wrote:
> 
>> IMO, that is the wrong way to look at it.
>> 
>> A users@ mailing list is a great, easy, low-cost and low-overhead way of
>> *increasing* the user community and providing an extra level of support.
>> Unless there is "strong evidence" that this is NOT the case, I would
>> recommend we create the list.
>> 
>>> On Jun 16, 2018, at 12:28 AM, Tianqi Chen 
>> wrote:
>>> 
>>> So unless there is a strong evidence that our community users prefers 
the
>>> mail-list, I would recommend we keep the current way
>>> 
>>> Tianqi
>>> 
>>> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández 
>> wrote:
>>> 
 Are we targeting just Seattle as our community? I really hope we are
 thinking a bit beyond that...
 
 On Fri, Jun 15, 2018, 21:22 Tianqi Chen 
>> wrote:
 
> I remember last time during the mxnet meetup in Seattle, we did a
>> survey,
> and most users preferred the current discuss forum. So I would say we
 stick
> with that given the user community prefers that
> 
> Tianqi
> 
> On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández 
> wrote:
> 
>> Then, if everybody agree, let's request the mailing list creation to
> INFRA
>> ;-)
>> 
>> Marco, I wouldn't do that. Typically developers are also subscribed
> there,
>> since they may be the most informed people for answering users'
> questions.
>> But the topics discussed there may not be of the interest for pure
>> development purposes. Some discussions will jump from users@ to dev@,
> but
>> at a different level. So I wouldn't forward one mailing list to the
> other.
>> 
>> On Fri, Jun 15, 2018, 21:01 Marco de Abreu
>>  wrote:
>> 
>>> I think nobody was opposed to it in the past, right?
>>> 
>>> I'd propose that all emails automatically get copied to dev@ to
 ensure
>>> high
>>> visibility initially. What do you think?
>>> 
>>> Sebastian  schrieb am Fr., 15. Juni 2018, 20:51:
>>> 
 I have already proposed this many times in the past and would
> strongly
 encourage it.
 
 -s
 
 On 15.06.2018 21:56, Sergio Fernández wrote:
> Hi,
> 
> is there any good reason why the podling doesn't have a users@
>> mailing
 list
> yet?
> 
> Honestly speaking, I'm not a big fan of the other tools the
 podling
>> is
> using. Slack and Web forums a cool tools, and I used them a lot
 in
>>> other
> contexts. But when it comes to transparency and community,
 mailing
>>> lists
> play a crucial role in the Apache Way.
> 
> Users are the most important asset a project can have. Even more
> than
> developers, believe me. So I think it's time to create a users@
>>> mailing
> list for to helping MXNet grow its community beyong the core
 team.
> 
> Cheers,
> 
 
>>> 
>> 
> 
 
>> 
>> 

Re: users@mxnet

[Jim Jagielski]

users@ mailing lists have great societal advantages that one shouldn't ignore...

And it's not like this is the only project with "multiple" communication 
choices for users. Most, if not all, projects have users@in addition to such 
supplemental methods as IRC channels, a forum, etc... It's about making it easy 
to have as many users as possible and as many potential ways for users to 
communicate. It's not confusing; it's empowering :)

> On Jun 18, 2018, at 1:19 PM, Tianqi Chen  wrote:
> 
> The problem of having multiple separate channels of communication is that
> users get confused, and the cost of maintenance goes up(people have to
> watch both). As the current community was at discuss forum and many users
> prefer it, having a mail-list is only a burden we will bring
> 
> Tianqi
> 
> On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski  wrote:
> 
>> IMO, that is the wrong way to look at it.
>> 
>> A users@ mailing list is a great, easy, low-cost and low-overhead way of
>> *increasing* the user community and providing an extra level of support.
>> Unless there is "strong evidence" that this is NOT the case, I would
>> recommend we create the list.
>> 
>>> On Jun 16, 2018, at 12:28 AM, Tianqi Chen 
>> wrote:
>>> 
>>> So unless there is a strong evidence that our community users prefers the
>>> mail-list, I would recommend we keep the current way
>>> 
>>> Tianqi
>>> 
>>> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández 
>> wrote:
>>> 
 Are we targeting just Seattle as our community? I really hope we are
 thinking a bit beyond that...
 
 On Fri, Jun 15, 2018, 21:22 Tianqi Chen 
>> wrote:
 
> I remember last time during the mxnet meetup in Seattle, we did a
>> survey,
> and most users preferred the current discuss forum. So I would say we
 stick
> with that given the user community prefers that
> 
> Tianqi
> 
> On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández 
> wrote:
> 
>> Then, if everybody agree, let's request the mailing list creation to
> INFRA
>> ;-)
>> 
>> Marco, I wouldn't do that. Typically developers are also subscribed
> there,
>> since they may be the most informed people for answering users'
> questions.
>> But the topics discussed there may not be of the interest for pure
>> development purposes. Some discussions will jump from users@ to dev@,
> but
>> at a different level. So I wouldn't forward one mailing list to the
> other.
>> 
>> On Fri, Jun 15, 2018, 21:01 Marco de Abreu
>>  wrote:
>> 
>>> I think nobody was opposed to it in the past, right?
>>> 
>>> I'd propose that all emails automatically get copied to dev@ to
 ensure
>>> high
>>> visibility initially. What do you think?
>>> 
>>> Sebastian  schrieb am Fr., 15. Juni 2018, 20:51:
>>> 
 I have already proposed this many times in the past and would
> strongly
 encourage it.
 
 -s
 
 On 15.06.2018 21:56, Sergio Fernández wrote:
> Hi,
> 
> is there any good reason why the podling doesn't have a users@
>> mailing
 list
> yet?
> 
> Honestly speaking, I'm not a big fan of the other tools the
 podling
>> is
> using. Slack and Web forums a cool tools, and I used them a lot
 in
>>> other
> contexts. But when it comes to transparency and community,
 mailing
>>> lists
> play a crucial role in the Apache Way.
> 
> Users are the most important asset a project can have. Even more
> than
> developers, believe me. So I think it's time to create a users@
>>> mailing
> list for to helping MXNet grow its community beyong the core
 team.
> 
> Cheers,
> 
 
>>> 
>> 
> 
 
>> 
>> 

Re: users@mxnet

[Tianqi Chen]

The problem of having multiple separate channels of communication is that
users get confused, and the cost of maintenance goes up(people have to
watch both). As the current community was at discuss forum and many users
prefer it, having a mail-list is only a burden we will bring

Tianqi

On Mon, Jun 18, 2018 at 9:48 AM, Jim Jagielski  wrote:

> IMO, that is the wrong way to look at it.
>
> A users@ mailing list is a great, easy, low-cost and low-overhead way of
> *increasing* the user community and providing an extra level of support.
> Unless there is "strong evidence" that this is NOT the case, I would
> recommend we create the list.
>
> > On Jun 16, 2018, at 12:28 AM, Tianqi Chen 
> wrote:
> >
> > So unless there is a strong evidence that our community users prefers the
> > mail-list, I would recommend we keep the current way
> >
> > Tianqi
> >
> > On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández 
> wrote:
> >
> >> Are we targeting just Seattle as our community? I really hope we are
> >> thinking a bit beyond that...
> >>
> >> On Fri, Jun 15, 2018, 21:22 Tianqi Chen 
> wrote:
> >>
> >>> I remember last time during the mxnet meetup in Seattle, we did a
> survey,
> >>> and most users preferred the current discuss forum. So I would say we
> >> stick
> >>> with that given the user community prefers that
> >>>
> >>> Tianqi
> >>>
> >>> On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández 
> >>> wrote:
> >>>
>  Then, if everybody agree, let's request the mailing list creation to
> >>> INFRA
>  ;-)
> 
>  Marco, I wouldn't do that. Typically developers are also subscribed
> >>> there,
>  since they may be the most informed people for answering users'
> >>> questions.
>  But the topics discussed there may not be of the interest for pure
>  development purposes. Some discussions will jump from users@ to dev@,
> >>> but
>  at a different level. So I wouldn't forward one mailing list to the
> >>> other.
> 
>  On Fri, Jun 15, 2018, 21:01 Marco de Abreu
>   wrote:
> 
> > I think nobody was opposed to it in the past, right?
> >
> > I'd propose that all emails automatically get copied to dev@ to
> >> ensure
> > high
> > visibility initially. What do you think?
> >
> > Sebastian  schrieb am Fr., 15. Juni 2018, 20:51:
> >
> >> I have already proposed this many times in the past and would
> >>> strongly
> >> encourage it.
> >>
> >> -s
> >>
> >> On 15.06.2018 21:56, Sergio Fernández wrote:
> >>> Hi,
> >>>
> >>> is there any good reason why the podling doesn't have a users@
>  mailing
> >> list
> >>> yet?
> >>>
> >>> Honestly speaking, I'm not a big fan of the other tools the
> >> podling
>  is
> >>> using. Slack and Web forums a cool tools, and I used them a lot
> >> in
> > other
> >>> contexts. But when it comes to transparency and community,
> >> mailing
> > lists
> >>> play a crucial role in the Apache Way.
> >>>
> >>> Users are the most important asset a project can have. Even more
> >>> than
> >>> developers, believe me. So I think it's time to create a users@
> > mailing
> >>> list for to helping MXNet grow its community beyong the core
> >> team.
> >>>
> >>> Cheers,
> >>>
> >>
> >
> 
> >>>
> >>
>
>

Re: users@mxnet

[Jim Jagielski]

IMO, that is the wrong way to look at it.

A users@ mailing list is a great, easy, low-cost and low-overhead way of 
*increasing* the user community and providing an extra level of support. Unless 
there is "strong evidence" that this is NOT the case, I would recommend we 
create the list.

> On Jun 16, 2018, at 12:28 AM, Tianqi Chen  wrote:
> 
> So unless there is a strong evidence that our community users prefers the
> mail-list, I would recommend we keep the current way
> 
> Tianqi
> 
> On Fri, Jun 15, 2018 at 9:25 PM, Sergio Fernández  wrote:
> 
>> Are we targeting just Seattle as our community? I really hope we are
>> thinking a bit beyond that...
>> 
>> On Fri, Jun 15, 2018, 21:22 Tianqi Chen  wrote:
>> 
>>> I remember last time during the mxnet meetup in Seattle, we did a survey,
>>> and most users preferred the current discuss forum. So I would say we
>> stick
>>> with that given the user community prefers that
>>> 
>>> Tianqi
>>> 
>>> On Fri, Jun 15, 2018 at 9:13 PM, Sergio Fernández 
>>> wrote:
>>> 
 Then, if everybody agree, let's request the mailing list creation to
>>> INFRA
 ;-)
 
 Marco, I wouldn't do that. Typically developers are also subscribed
>>> there,
 since they may be the most informed people for answering users'
>>> questions.
 But the topics discussed there may not be of the interest for pure
 development purposes. Some discussions will jump from users@ to dev@,
>>> but
 at a different level. So I wouldn't forward one mailing list to the
>>> other.
 
 On Fri, Jun 15, 2018, 21:01 Marco de Abreu
  wrote:
 
> I think nobody was opposed to it in the past, right?
> 
> I'd propose that all emails automatically get copied to dev@ to
>> ensure
> high
> visibility initially. What do you think?
> 
> Sebastian  schrieb am Fr., 15. Juni 2018, 20:51:
> 
>> I have already proposed this many times in the past and would
>>> strongly
>> encourage it.
>> 
>> -s
>> 
>> On 15.06.2018 21:56, Sergio Fernández wrote:
>>> Hi,
>>> 
>>> is there any good reason why the podling doesn't have a users@
 mailing
>> list
>>> yet?
>>> 
>>> Honestly speaking, I'm not a big fan of the other tools the
>> podling
 is
>>> using. Slack and Web forums a cool tools, and I used them a lot
>> in
> other
>>> contexts. But when it comes to transparency and community,
>> mailing
> lists
>>> play a crucial role in the Apache Way.
>>> 
>>> Users are the most important asset a project can have. Even more
>>> than
>>> developers, believe me. So I think it's time to create a users@
> mailing
>>> list for to helping MXNet grow its community beyong the core
>> team.
>>> 
>>> Cheers,
>>> 
>> 
> 
 
>>> 
>> 

Re: Nightly tests README accurate?

[Meghna Baijal]

Hi,
This readme is incorrect and I am fixing this as a part of the nightly
tests PR soon.

Thanks,
Meghna

On Sun, Jun 17, 2018 at 7:05 AM kellen sunderland <
kellen.sunderl...@gmail.com> wrote:

> I'm not a commiter, but I would say this is clearly not correct.  I would
> propose removing the doc until someone has time to verify that the steps
> work correctly, and that the descriptions are accurate.
>
> On Thu, Jun 14, 2018 at 10:04 PM Indhu  wrote:
>
> > Is the README
> > 
> for
> > the nightly tests accurate? For example,
> >
> > 1. Are tests being run on machines with Intel i7-4790 and 4 Nvidia GTX
> 970
> > Tis?
> > 2. Is http://ci.dmlc.ml/ the right place to look for build status?
> > 3. Is the instruction to run on Jenkins correct?
> >
> > If not, what all needs to be changed in that page?
> >
>