Description
I package Python packages as wheels using the pip wheel command. Scikit-learn does not report its requirements to setuptools as install_requires or setup_requires. This makes pip wheel fail.
Description
The current checks for the presence of numpy and scipy during installation is according to what I found in the repo and the tracker based on Issue #1495 and PR #4371.
I suggest to avoid raising errors in setup.py and list numpy as an setup requirement and numpy and scipy as an install requirement, so that pip/wheel can install the dependencies seamlessly in the background, if they are not present. Pip's setup_requires section should list all packages necessary to invoke setup.py subbcommands (http://setuptools.readthedocs.io/en/latest/setuptools.html#setup_requires).
In my usage example, pip wheel would install numpy then temporarily for the creation of the scikit-learn wheel.
Steps/Code to Reproduce
virtualenv sklearntest
. sklearntest/bin/activate
pip install -U pip setuptools wheel
pip wheel --no-binary :all: scikit-learn
Expected Results
A scikit-learn wheel should be buildable as simple as
% pip install scikit-learn
Processing /whatever-dir/devel/scikit-learn
Collecting numpy (from scikit-learn==0.19.dev0)
File was already downloaded /whatever-dir/devel/scikit-learn/numpy-1.11.2-cp27-cp27m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Collecting scipy (from scikit-learn==0.19.dev0)
File was already downloaded /whatever-dir/devel/scikit-learn/scipy-0.18.1-cp27-cp27m-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
[..............]
Successfully built scikit-learn
pip wheel . 87.87s user 9.66s system 76% cpu 2:07.13 total
Actual Results
Collecting scikit-learn
Downloading scikit-learn-0.18.1.tar.gz (8.9MB)
100% |################################| 8.9MB 143kB/s
Building wheels for collected packages: scikit-learn
Running setup.py bdist_wheel for scikit-learn ... error
Complete output from command /private/tmp/sklearntest/bin/python2.7 -u -c "import setuptools, tokenize;__file__='/private/var/folders/kn/3fznwwr15z9chdd30p3q4qm80000gp/T/pip-build-489cEa/scikit-learn/setup.py';f=getattr(tokenize, 'open', open)(__file__);code=f.read().replace('\r\n', '\n');f.close();exec(compile(code, __file__, 'exec'))" bdist_wheel -d /var/folders/kn/3fznwwr15z9chdd30p3q4qm80000gp/T/tmpMCqZC2pip-wheel-:
Partial import of sklearn during the build process.
Traceback (most recent call last):
File "/private/var/folders/kn/3fznwwr15z9chdd30p3q4qm80000gp/T/pip-build-489cEa/scikit-learn/setup.py", line 169, in get_numpy_status
import numpy
ImportError: No module named numpy
Traceback (most recent call last):
File "/private/var/folders/kn/3fznwwr15z9chdd30p3q4qm80000gp/T/pip-build-489cEa/scikit-learn/setup.py", line 149, in get_scipy_status
import scipy
ImportError: No module named scipy
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "/private/var/folders/kn/3fznwwr15z9chdd30p3q4qm80000gp/T/pip-build-489cEa/scikit-learn/setup.py", line 270, in <module>
setup_package()
File "/private/var/folders/kn/3fznwwr15z9chdd30p3q4qm80000gp/T/pip-build-489cEa/scikit-learn/setup.py", line 250, in setup_package
.format(numpy_req_str, instructions))
ImportError: Numerical Python (NumPy) is not installed.
scikit-learn requires NumPy >= 1.6.1.
Installation instructions are available on the scikit-learn website: http://scikit-learn.org/stable/install.html
Discussion
The nature of my request is basically, to make scikit-learn behave more closely to how other packages behave.