Conda about treelite HOT 14 CLOSED

The Conda package is available now: https://anaconda.org/conda-forge/treelite

from treelite.

I submitted a Conda recipe to conda-forge today: conda-forge/staged-recipes#11926.

from treelite.

+1

from treelite.

+1

from treelite.

+1

from treelite.

• 1

from treelite.

+1

from treelite.

+1

from treelite.

@hcho3 How hard would it actually be to provide a conda installer for treelite?
@teju85 @dantegd FYI.

from treelite.

@canonizer I'm currently working on 1.0 release for PyPI. What is your use case for using Conda specifically?

from treelite.

Most rapids AI components don't support PyPI due to c++14

from treelite.

@trivialfis Thanks for clarification. I have no prior experience setting up a conda recipe. I wonder how difficult it is.

from treelite.

Would suggest adding this to conda-forge. Should add I'm biased as I work on the project. That said, conda-forge has become the defacto community Conda channel. It is also used to build packages in the Anaconda distribution. Many other domain specific channels (like rapidsai) use conda-forge as their base.

When it comes to contributing a new recipe, we have a quickstart for contributing and a FAQ. For more comprehensive information conda-forge has docs and conda-build has even more docs. Plus there is an active Gitter channel and there are GitHub teams like the staged-recipes team designed to walk you through the process.

Please let me know if you have any questions. 😄

from treelite.

@hcho3 the process is the following (based on https://conda-forge.org/#add_recipe):

• fork the repository https://github.com/conda-forge/staged-recipes
• create treelite folder in recipes
• create meta.yaml with the package description and dependencies
• create build.sh and bld.bat to build on unix and windows.

Here is an attempt of conda-build for treelite: conda-forge/staged-recipes#11338.

Here are some remarks:

• the tar.gz archive for release 0.32 is missing dmlc-core source so I had to use git_url instead of source tag in the meta.yaml
• openmp support in conda is explained here: https://conda-forge.org/docs/maintainer/knowledge_base.html#openmp
• on osx, I had patch your CMakeLists.txt to include OpenMP headers (based on: https://cmake.org/cmake/help/latest/module/FindOpenMP.html).
• It might be better to build dmlc-core separately and include it in treelite CMake using exported targets, this way you could use dmlc package which is already available on conda.

from treelite.