search

poke1024 / origami

A suite of batches and tools for OCR tasks.
71 stars 16 forks source link

scikit-geometry is only available through conda-forge #17

Open diegosiqueir4 opened 3 years ago

diegosiqueir4 commented 3 years ago

I tried to install all requirements using the following pip commands (as recommended in #9):

pip install -v -r requirements/pip.txt 
pip install -v -r requirements/conda.txt

However, the scikit-geometry package could not be found:

ERROR: Could not find a version that satisfies the requirement scikit-geometry (from versions: none)
ERROR: No matching distribution found for scikit-geometry

According to this issue it's not yet possible to install scikit-geometry from pypi. The package is only available through conda-forge.

Thought it might be useful to share this information with you.

bertsky commented 3 years ago

That particular tail is devilishly long:

If you want to build scikit-geometry from source, you need a system library called libcgal-dev – but at least version 5 IIUC.

However, installing CGAL from source does not work either, unless you have cmake>=3.11.

If you still manage, compilation of skgeom will take 4 GB RSS and ages to finish.

EDIT: Forgot to mention that the package name in requirements/conda.txt is therefore inconsistent with the setup name.

poke1024 commented 3 years ago

@bertsky I really should trash requirements/* in favor of one central conda-based environment.yml, which then relies on https://anaconda.org/conda-forge/scikit-geometry for scikit.

bertsky commented 3 years ago

I really should trash requirements/* in favor of one central conda-based environment.yml, which then relies on https://anaconda.org/conda-forge/scikit-geometry for scikit.

Please don't remove these lists without some other mechanism for pip – it's the only way to get a non-conda deployment. (I am in the middle of building an OCR-D wrapper for Origami and would like to get it into ocrd_all, where conda is more or less ruled out.)

poke1024 commented 3 years ago

Please don't remove these lists without some other mechanism for pip – it's the only way to get a non-conda deployment. (I am in the middle of building an OCR-D wrapper for Origami and would like to get it into ocrd_all, where conda is more or less ruled out.)

ok, I see.