Mailman 3 June 2018 - scikit-image

Re: [scikit-image] Should pyamg be an optional dependency?
by Mark Harfouche 29 Jun '18

29 Jun '18

Thanks guys. That clears things up. Regarding the installation, it seems to be pretty ok now. Maybe that was before wheels were super popular. I came across it as I was going through the build process again finalising the removal of matplotlib as a dependency. That is mostly done, but the PR turned out to be more complicated than I thought so I split it in two. On Thu, Jun 28, 2018 at 7:10 PM Josh Warner <silvertrumpet999(a)gmail.com> wrote: > The only algorithm I'm aware that it enhances is random walker > segmentation. That said, it's of massive benefit to random walker. > > It is truly an optional dependency in that the algorithm transparently > functions without it. > > From a test matrix standpoint I'm not sure this warrants more than > including it with current latest stable dependencies - but I believe there > was once some difficulty getting it up and running in an efficient manner > on Travis. > > > On Thu, Jun 28, 2018, 13:06 Mark Harfouche <mark.harfouche(a)gmail.com> > wrote: > >> Hi all, >> >> Should pyamg just be listed as an optional dependency? >> As far as i know, it is meant to speed up some scipy algorithms. >> Is there a reason it is not in there? >> >> Mark >> > _______________________________________________ >> scikit-image mailing list >> scikit-image(a)python.org >> https://mail.python.org/mailman/listinfo/scikit-image >> > _______________________________________________ > scikit-image mailing list > scikit-image(a)python.org > https://mail.python.org/mailman/listinfo/scikit-image >

1 0

Re: [scikit-image] Should pyamg be an optional dependency?
by Josh Warner 29 Jun '18

29 Jun '18

The only algorithm I'm aware that it enhances is random walker segmentation. That said, it's of massive benefit to random walker. It is truly an optional dependency in that the algorithm transparently functions without it. >From a test matrix standpoint I'm not sure this warrants more than including it with current latest stable dependencies - but I believe there was once some difficulty getting it up and running in an efficient manner on Travis. On Thu, Jun 28, 2018, 13:06 Mark Harfouche <mark.harfouche(a)gmail.com> wrote: > Hi all, > > Should pyamg just be listed as an optional dependency? > As far as i know, it is meant to speed up some scipy algorithms. > Is there a reason it is not in there? > > Mark > _______________________________________________ > scikit-image mailing list > scikit-image(a)python.org > https://mail.python.org/mailman/listinfo/scikit-image >

1 0

Re: [scikit-image] Should pyamg be an optional dependency?
by Stefan van der Walt 29 Jun '18

29 Jun '18

On Thu, 28 Jun 2018 13:06:38 -0700, Mark Harfouche wrote: > Should pyamg just be listed as an optional dependency? > As far as i know, it is meant to speed up some scipy algorithms. > Is there a reason it is not in there? pyamg is an optional dependency for the random walker solver, correct. Stéfan

1 0

Should pyamg be an optional dependency?
by Mark Harfouche 28 Jun '18

28 Jun '18

Hi all, Should pyamg just be listed as an optional dependency? As far as i know, it is meant to speed up some scipy algorithms. Is there a reason it is not in there? Mark

1 0

Re: [scikit-image] Are windows builds on appveyor broken?
by Mark Harfouche 27 Jun '18

27 Jun '18

I think it is because the scipy --pre builds are broken. I'm working on a pr that should fix it. On Tue, Jun 26, 2018 at 7:26 PM Juan Nunez-Iglesias <jni(a)fastmail.com> wrote: > I've noticed a couple of similar failures, with a missing DLL error. I > don't know what's causing them. > > > On Wed, Jun 27, 2018, at 10:59 AM, Mark Harfouche wrote: > > I just triggered a build on master, and Appveyor is failing. Any clue why? > > https://ci.appveyor.com/project/scikit-image/scikit-image/history > > > *_______________________________________________* > scikit-image mailing list > scikit-image(a)python.org > https://mail.python.org/mailman/listinfo/scikit-image > > > _______________________________________________ > scikit-image mailing list > scikit-image(a)python.org > https://mail.python.org/mailman/listinfo/scikit-image >

1 0

Re: [scikit-image] Are windows builds on appveyor broken?
by Juan Nunez-Iglesias 27 Jun '18

27 Jun '18

I've noticed a couple of similar failures, with a missing DLL error. I don't know what's causing them. On Wed, Jun 27, 2018, at 10:59 AM, Mark Harfouche wrote: > I just triggered a build on master, and Appveyor is failing. Any > clue why?> > https://ci.appveyor.com/project/scikit-image/scikit-image/history > > > _________________________________________________ > scikit-image mailing list > scikit-image(a)python.org > https://mail.python.org/mailman/listinfo/scikit-image

1 0

Re: [scikit-image] Minimum Python version?
by Thomas Caswell 27 Jun '18

27 Jun '18

For Matplotlib we are going with 3 years worth of python versions ( https://matplotlib.org/devdocs/devel/min_dep_policy.html ) which seems like a reasonable compromise between keeping the library up-to-date and keep users stuck on old(er) versions of python able to access master. We also thought about "last 2 versions of python" but if you plan to release near a python release date things get a bit dicey planning wise. Tom On Tue, Jun 26, 2018 at 9:18 PM Juan Nunez-Iglesias <jni(a)fastmail.com> wrote: > > On Wed, Jun 27, 2018, at 9:03 AM, Stefan van der Walt wrote: > > $ pypinfo skimage pyversion > > pypinfo! Awesome! > > > | python_version | download_count | > > | -------------- | -------------- | > > | 3.6 | 3,172 | > > | 2.7 | 1,995 | > > | 3.5 | 1,866 | > > | 3.4 | 118 | > > | None | 85 | > > | 3.7 | 18 | > > > > It looks like 3.5 currently makes up about 37% of Python 3 installs. > > This number should come down over the next few months, but I suggest we > > keep supporting 3.5 for at least one more release. > > Another way of looking at these numbers is that 3.5 makes up a smaller > proportion than 2.7, which we are now supporting only via the LTS. ;) But I > am being facetious. The sum of 2.7 and 3.5 is more than 3.6. It would be > bad to only support a minority of our users in master. > > Thanks for the info! I agree, let's wait until one month after 0.15 and > check the stats again at that point. > > Juan. > > ... > > Ok so I looked at the command and thought, hmm, should that be > scikit-image? After 30 painful minutes of setting up my own bigquery > credentials: > > pypinfo scikit-image pyversion > > | python_version | download_count | > | -------------- | -------------- | > | 2.7 | 61,274 | > | 3.6 | 23,767 | > | 3.5 | 20,466 | > | None | 3,693 | > | 3.4 | 1,048 | > | 3.7 | 81 | > | 3.3 | 3 | > | 2.6 | 2 | > | 3.2 | 1 | > > This comes with all the caveats of tracking PyPI downloads (most are from > CI etc), but nevertheless: yes, let's keep 3.5 around. And let's make good > on our LTS promise. =) > _______________________________________________ > scikit-image mailing list > scikit-image(a)python.org > https://mail.python.org/mailman/listinfo/scikit-image >

1 0

Re: [scikit-image] Minimum Python version?
by Stefan van der Walt 27 Jun '18

27 Jun '18

On Tue, 26 Jun 2018 17:11:25 +1000, Juan Nunez-Iglesias wrote: > What Python version do we want for 0.15? Currently, it is set to 3.5. I > want to make a motion for 3.6. This gives us: > - f-strings, which is super nice syntax. > - dictionaries maintain insertion order, which is also super nice. (We > would be able to replace all OrderedDict instances with plain > dicts.) Features are nice (very nice, in this case) to have, but in itself is not a good enough reason to switch. Let's look at our installation statistics: $ pypinfo skimage pyversion | python_version | download_count | | -------------- | -------------- | | 3.6 | 3,172 | | 2.7 | 1,995 | | 3.5 | 1,866 | | 3.4 | 118 | | None | 85 | | 3.7 | 18 | It looks like 3.5 currently makes up about 37% of Python 3 installs. This number should come down over the next few months, but I suggest we keep supporting 3.5 for at least one more release. Also, note the 18 installs on 3.7. We should get ready! Best regards, Stéfan

2 1

Are windows builds on appveyor broken?
by Mark Harfouche 27 Jun '18

27 Jun '18

I just triggered a build on master, and Appveyor is failing. Any clue why? https://ci.appveyor.com/project/scikit-image/scikit-image/history

1 0

Re: [scikit-image] Minimum Python version?
by François Boulogne 26 Jun '18

26 Jun '18

Le 26/06/2018 à 09:11, Juan Nunez-Iglesias a écrit : > Hi all, > > What Python version do we want for 0.15? Currently, it is set to 3.5. > I want to make a motion for 3.6. This gives us: > > - f-strings, which is super nice syntax. > - dictionaries maintain insertion order, which is also super nice. (We > would be able to replace all OrderedDict instances with plain dicts.) > > I certainly want 3.6+ for skimage 1.0. The question is whether to pull > the trigger now or later. > Very good question Juan. I'm someone who tend to push to the bleeding edge for development. My opinion is that being conservative tends to slow down everyone sooner or later. To be more rational: * Debian stable version (March '18), py3.5: https://packages.debian.org/stable/python/python3 * Debian testing (next stable version), py3.6: https://packages.debian.org/testing/python/python3 Pro to keep v3.5: * It is likely that our next release will happen with debian stable providing py3.5, so our release won't be pip installable. Con: but those people have the LTS version and they have a deb package for skimage 0.12 https://packages.debian.org/stretch/python3-skimage my 2 cents. -- François Boulogne. http://www.sciunto.org GPG: 32D5F22F

1 0