Xgboost: PyPi installed xgboost==0.4a30 yields Mac OS X library on Linux platforms

Created on 25 May 2016 · 6Comments · Source: dmlc/xgboost

xgboost==0.4a30

Operating System: Ubuntu 14.04 LTS

python -c 'import xgboost'
OSError: /usr/local/lib/python2.7/dist-packages/xgboost/./wrapper/libxgboostwrapper.so: invalid ELF header

file /usr/local/lib/python2.7/dist-packages/xgboost/./wrapper/libxgboostwrapper.so
/usr/local/lib/python2.7/dist-packages/xgboost/./wrapper/libxgboostwrapper.so: Mach-O 64-bit x86_64 dynamically linked shared library

Error is also present when passing the --no-binary :all: option to pip install

pip install --no-cache-dir --no-binary :all: xgboost

Source

moandcompany

👍4 😄1

Most helpful comment

The current workaround is using the normal installation route -- build from source.

For Linux users:

git clone --recursive https://github.com/dmlc/xgboost
cd xgboost; make -j4
sudo python python-package/setup.py install

To install the latest tagged version (0.47) from source, check out that branch beforehand:

git clone --recursive https://github.com/dmlc/xgboost
cd xgboost
git checkout tags/0.47
make -j4
sudo python python-package/setup.py install

moandcompany on 25 May 2016

🎉3

All 6 comments

The current workaround is using the normal installation route -- build from source.

For Linux users:

git clone --recursive https://github.com/dmlc/xgboost
cd xgboost; make -j4
sudo python python-package/setup.py install

To install the latest tagged version (0.47) from source, check out that branch beforehand:

git clone --recursive https://github.com/dmlc/xgboost
cd xgboost
git checkout tags/0.47
make -j4
sudo python python-package/setup.py install

moandcompany on 25 May 2016

🎉3

seems like the compilation failed on your system. please make sure you have gcc on your system (by apt-get install build-essential on ubuntu, other platform please change)

the reason: the libxgboostwrapper.so is a place holder for packing for pip which should be automatically deleted after successful installation.

surely, installing from github as @moandcompany said works too, and it is the latest version with good new features.

phunterlau on 26 May 2016

The Python package scikit-learn is also a prequisite dependency that is not automatically pulled in during the xgboost setup.py install process via pip

moandcompany on 26 May 2016

We also observed that it appears the package version number is not being incremented on the 0.47 tag

python -c 'import xgboost; xgboost.__version__' appears to report version 0.4

moandcompany on 27 May 2016

That is true that pip version is outdated. The new refactored code doesn't follow the old way for compiling on the fly when use pip, and some tries on deploying the new code but some troubles, so I rolled back and kept the old version on PyPI for now, will soon fix it.

phunterlau on 27 May 2016

Hi,
this issue is still present on Debian 9, pip 9.0.1, Python 2.7.13 and xgboost-0.6a2.