Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Tom de Geus
    @tdegeus
    Also, I don't completely follow the logic: the fact that indices start at zero already takes care of this, in the case the that shape(d) == 0 one cannot specify i != 0, hence the formula would work just fine
    Johan Mabille
    @JohanMabille
    there is broadcasting
    if you have a + b with a.shape = (2, 1) and b.shape = (2, 2)
    it is fine to ask for (a + b)(1, 1)
    which will transform into a(1, 1) + b(1, 1)
    you need a 0 stride to cancel the 1 index
    ndbecker
    @nbecker
    tried installing xtl, xsimd, and xtensor using cmake, but still get tons of errors compiling, starting with 'XSIMD_DEFAULT_ALIGNMENT' was not declared in this scope
    Johan Mabille
    @JohanMabille
    so, as I said previously, the current master / xsmid 7.5.1 is not compatible with xtensor
    you ned to checkout version 7.5.0
    ndbecker
    @nbecker
    ok
    Johan Mabille
    @JohanMabille
    and if you are on Windows with VS2019, it won't build (I need to release a working 7.5.1)
    ndbecker
    @nbecker
    oh no windows here!
    Tom de Geus
    @tdegeus
    Fair enough. Though one might argue that the public strides should not have to know about internal broadcasting solutions.
    Johan Mabille
    @JohanMabille
    it does not, broadcasting can be "external"
    i.e x = xt::broadcast(a, xshape(2, 2))(1, 1) should be equivalent to a(1, 1)
    meaning you can access a singl broadcasted value
    Tom de Geus
    @tdegeus
    ok fair enough
    the opposition rests its case ;)
    So I proposed a fix for argsort : xtensor-stack/xtensor#2238
    Loic Gouarin
    @gouarin
    A colleague asked me if it was possible to use matplotlib with xeus-cling. I thought about it and I have an idea which (I think) could be easy to implement.
    You probably know that a package called matplotlib-cpp exists (https://github.com/lava/matplotlib-cpp) that makes the job using CPython but not into a notebook unfortunately.
    But since xeus-python is already in C++ and uses pybind11, I think we could implement a matplotlib wrapper using pybind11 to manage C++ containers and use them with matplotlib inside xeus-python by using its interpreter.
    What do you think about the idea ?
    A question remains: is it possible to use matplotlib with xeus-python and have the magic %matplotlib ?
    I didn't see the magics in the documentation
    Sylvain Corlay
    @SylvainCorlay

    A question remains: is it possible to use matplotlib with xeus-python and have the magic %matplotlib ?

    Yes, but not with the qt backend.

    Loic Gouarin
    @gouarin
    is it an issue ?
    Sylvain Corlay
    @SylvainCorlay
    Not in the notebook.

    But since xeus-python is already in C++ and uses pybind11, I think we could implement a matplotlib wrapper using pybind11 to manage C++ containers and use them with matplotlib inside xeus-python by using its interpreter. What do you think about the idea ?

    Yep, I guess that it would work - but in the end you still need python.

    Loic Gouarin
    @gouarin
    Yes but for now, there is no equivalent in c++
    I don't see where the magic matplotlib is implemented. I thought it was part of the kernel
    Loic Gouarin
    @gouarin
    The magics are in IPython. I understand now ;)
    Johan Mabille
    @JohanMabille
    @nunupeke:matrix.org I cannot reproduce your issue with xsimd on VS2019
    Johan Mabille
    @JohanMabille
    can you open an issue on xsimd with your complete setup and the error message you have? Thanks!
    nunupeke
    @nunupeke:matrix.org
    [m]
    Did you test against v7.5.0? Actually I use VS2017 (MSVC 19...) - damn Visual Studio version numbers! Is VS2019 recommended?
    Johan Mabille
    @JohanMabille
    VS2015, 2017 and 2019 all work
    nunupeke
    @nunupeke:matrix.org
    [m]
    Hm, ok. Will open an issue then. Thanks again!
    Johan Mabille
    @JohanMabille
    you're welcome ;) And yes I've tested against 7.5.0
    Sylvain Corlay
    @SylvainCorlay
    ipycytoscape moving to the cytoscape consortium: https://twitter.com/mari_meir/status/1392052543447085056
    Marianne Corvellec
    @mkcor
    Hello voilà people, especially voila-reveal people @SylvainCorlay @maartenbreddels @martinRenou! I'm co-presenting at ImageXD so I used voila-reveal to make a slideshow (our own dog food, hehe). I noticed that voilà had grown even stronger security-wise and the docs are always very neat! But (using --template=reveal) I couldn't get the actual slideshow (separate slides and navigation arrows to change slides) unless I would add roughly this piece: voila-dashboards/voila-reveal@957dffb somewhere in the templates (whether under PREFIX/share/jupyter/nbconvert/templates/reveal/ or under PREFIX/share/jupyter/voila/templates/reveal/). Adding it coarsely (say, at the top of the existing template files) isn't satisfactory though, because it overwrites all the styling otherwise perfectly rendered by voilà. :eyes:
    Sylvain Corlay
    @SylvainCorlay
    Hey @mkcor are you using some old version of nbconvert?
    Marianne Corvellec
    @mkcor
    @SylvainCorlay oh, let me check... nbconvert==5.6.1
    By now I've hacked my envs and the templates files so much that I should start again from a clean env.
    Sylvain Corlay
    @SylvainCorlay
    voila 0.2 and the latest voila-reveal requires nbconvert 6
    Marianne Corvellec
    @mkcor
    aaaah ok! thanks :)
    I'll start again 100% mamba (which wasn't the case, my collaborator having set up the project with venv/pip) ;)
    Sylvain Corlay
    @SylvainCorlay
    you can start from mambaforge.
    Marianne Corvellec
    @mkcor
    oh, is it new?
    README says mamba install -c conda-forge voila-reveal
    Sylvain Corlay
    @SylvainCorlay
    mambaforge is like miniconda with mamba in the base environment, which defaults to conda-forge.
    it's da best.
    Marianne Corvellec
    @mkcor
    aaah ok :D
    Marianne Corvellec
    @mkcor
    starred :wink: :ok_hand: