Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • 13:24
    rfezzani commented #3892
  • 13:20
    rfezzani labeled #5029
  • 13:20
    rfezzani opened #5029
  • 08:25
    alexjarosch commented #5026
  • 06:00
    sciunto edited #5028
  • 05:59
    alexjarosch commented #5026
  • 05:58
    sciunto labeled #5028
  • 05:58
    sciunto opened #5028
  • 05:58
    sciunto labeled #5028
  • 05:55
    alexjarosch commented #5026
  • 04:56
    stefanv commented #5026
  • 04:23
    haesleinhuepf commented #5006
  • 04:10
    jni commented #5006
  • 04:09
    jni commented #5006
  • 04:08
    jni commented #5006
  • 03:54
    haesleinhuepf commented #5006
  • 03:45
    stefanv commented #47
  • 03:45

    stefanv on master

    Add config to build aarch64 whe… Merge pull request #47 from jan… (compare)

Matt Wenham
@MattWenham
@hmaarrfk I hear you...
Mark Harfouche
@hmaarrfk
I think this is especially true in windows, whereby you activate by opening the anaconda prompt
i think you may be able to open the prompt of a specific environment directly
Juan Nunez-Iglesias
@jni
@MattWenham d'oh! That's very interesting, I wonder if I was having similar issues when I last encountered this error message. Looks like we are going to have to start recommending python -m pytest instead of pytest, just like we do with pip now...
also, I had always felt like base should be pristine, but never really had a good reason why. Thanks for giving me a concrete example! =)
As to your np.round question, I have no idea, I presume because some of the operations are quite sensitive to approximation errors, but I still think that should be reflected in the assert_allclose call rather than by calling np.round. Usually if something is really bugging me I use git blame to try to track down the PR where a change happened and see whether any of the discussion there can enlighten me as to the motivation for a change
more often than not, it's just that nobody thought to question a particular design choice at the time.
Kira Evans
@kne42
does this bug persist when you use the new shell-based conda where you have to explicitly conda activate on startup (which activates the conda root environment)?
Juan Nunez-Iglesias
@jni
@kne42 I thought possibly not but actually, I just checked, the base environment is not in my path in the new conda, so possibly it disappears!
Matt Wenham
@MattWenham
Thanks for your thoughts, folks!
I'm now happy with the code related to scikit-image/scikit-image#3414, so how do I submit it for review?
Kira Evans
@kne42
I mean as long as your PR is open and up to date, you just wait for a team member to get around to reviewing it
MattWenham @MattWenham twiddles thumbs
Juan Nunez-Iglesias
@jni
:joy: hold, please! =)
Matt Wenham
@MattWenham
Is my call important to you? 😂😂😂
seansabe
@seansabe
Greetings.
I am trying to install the scikit-image through conda install but it sticks forever and nothing happens.
¿Any idea?
Donwload isn't working either.
Download*
Benjamin Root
@WeatherGod
try doing a conda update conda. Some versions of conda had bad dependency solvers and would spend forever figuring out the dependency chains
seansabe
@seansabe
I did it but still sticking.
Juan Nunez-Iglesias
@jni
@seansabe very weird. are you installing from conda-forge or defaults?
also, are you able to conda-install other things?
seansabe
@seansabe
@jni Yes, I am using conda-forge. And I am able to install other stuff.
Mark Harfouche
@hmaarrfk
presently, the challenge is likely that scikit-image depends on matplotlib, which depends on qt
it takes time for my environment to solve
Benjamin Root
@WeatherGod
yeah, we just recently noticed some dependency issues with matplotlib+qt on conda yesterday. I don't think we figured out exactly what's wrong, but it might be qtconsole that is the root of the issues
Juan Nunez-Iglesias
@jni
@WeatherGod did something install eventually? @seansabe are you able to create a brand new environment? conda create --name skimage-test --channel=conda-forge python=3.7 scikit-image=0.14.1
Matt Wenham
@MattWenham
@jni - just looking through your revisions on my PR. What's missing from the docstring for _coprop_weights? Is it just the Returns that are missing or something more?
Matt Wenham
@MattWenham
@jni - Would you expect all my helper functions to have full docstrings? Can / how can I make them private to the texture Cython module??
Matt Wenham
@MattWenham
Ignore the 'private' bit, I've got it now...
Matt Wenham
@MattWenham
It's nearly there... could someone please investigate why one of the automated builds is failing? I don't think this has anything to do with my changes 🤔
Matt Wenham
@MattWenham
It's failing on both builds which use PIP_FLAGS="--pre" so I assume it's something to do with a pre-release version of another library.
Juan Nunez-Iglesias
@jni
@MattWenham you are correct, those are not your fault.
Ghost
@ghost~565b49ee16b6c7089cbc9b44
Hi! :) Are there any libraries to read/write binary .bmp files in Python ecosystem?
Davis Bennett
@d-v-b
Ghost
@ghost~565b49ee16b6c7089cbc9b44
@d-v-b imageio doesn't work with booleans, as far as I remember.
Davis Bennett
@d-v-b
ah sorry, i misread "binary"
Mark Harfouche
@hmaarrfk
@soupault you can try to add the functionality you need https://github.com/hmaarrfk/redpil
Juan Nunez-Iglesias
@jni
@soupault my very vague understanding of these things is that bmp is a pretty simple format. Perhaps you should extend imageio to support binary bmps? =)
Mark Harfouche
@hmaarrfk
@jni, yeah redpil does alot of reading the metadata and builds the infrastructure you would need to support these images.
I actually think I implemented reading boolean images. I have a test for it:
https://github.com/hmaarrfk/redpil/blob/master/redpil/tests/test_bmp.py#L76
The hard thing to understand is that boolean images are forced to be palletted, so even though there are only 2 possible values, those values will have RGB values assigned to them.
but the test also seems to indicated that pillow is also able to read the same image
Juan Nunez-Iglesias
@jni
@hmaarrfk why did you decide to write redpil independently rather than contribute to imageio?
Mark Harfouche
@hmaarrfk
just faster development cycle. imageio also stated that it was a frontend to other plugins
Mark Harfouche
@hmaarrfk
eventually, i guess I didn't care for a few reasons:
  1. bitmaps force your data to be 4 byte aligned. Seems like an unecessary limitation for RGB or gray data.
  2. imageio got the pillow settings correct so as to write images at (native?) speeds
Juan Nunez-Iglesias
@jni
@hmaarrfk what do you mean you didn't care? Are you saying the redpil bmp code is bit-rotting/unmaintained? I don't have much skin in this game, just want to get an accurate view of the image io landscape in Python
Mark Harfouche
@hmaarrfk
I think to displace Pillow would take a lot of work. Putting some experimental code in imageio would be dangerous. Originally, i thought there would be a speed advantage. And there was, but I submitted some patches to imageio that helped in that regard. Now redpil supports a different subset of bmp than Pillow so that might be a plus. There were a few examples in some BMP library that I wanted to get to really complete coverage of the BMP standard.