Develop Reference

Onnx installation issue on linux

I tried to install onnx on my linux machine and encountered issue when I use python3 -m pip install onnxas
/tmp/pip-install-7ajubmjo/onnx_08cda1ef5f1d460ab7b455210cf3083e/onnx/defs/sequence/defs.cc: In function ‘bool onnx::BuildSequenceMapBodyFunc(const onnx::FunctionBodyBuildContext&, const onnx::OpSchema&, onnx::FunctionProto&)’: /tmp/pip-install-7ajubmjo/onnx_08cda1ef5f1d460ab7b455210cf3083e/onnx/defs/sequence/defs.cc:675:40: error: no match for ‘operator[]’ (operand types are ‘google::protobuf::RepeatedPtrField<onnx::ValueInfoProto>’ and ‘int’)
seq_at_node.add_output(g_inputs[inputIndex].name());
[ 95%] Building CXX object CMakeFiles/onnx.dir/onnx/version_converter/helper.cc.o
CMakeFiles/onnx.dir/build.make:830: recipe for target 'CMakeFiles/onnx.dir/onnx/defs/sequence/defs.cc.o' failed
The protobuf was installed via sudo apt-get install protobuf-compiler libprotobuf-dev and my environment is:
- python=3.6
- ubuntu=18.04
Another related issue seems that the compilation cannot not find pybind11 as:
Generated: /tmp/pip-install-7ajubmjo/onnx_08cda1ef5f1d460ab7b455210cf3083e/.setuptools-cmake-build/onnx/onnx-ml.proto
Generated: /tmp/pip-install-7ajubmjo/onnx_08cda1ef5f1d460ab7b455210cf3083e/.setuptools-cmake-build/onnx/onnx-operators-ml.proto
Generated: /tmp/pip-install-7ajubmjo/onnx_08cda1ef5f1d460ab7b455210cf3083e/.setuptools-cmake-build/onnx/onnx-data.proto
-- Could NOT find pybind11 (missing: pybind11_DIR)
-- pybind11 v2.9.1
--
-- ******** Summary ********
-- CMake version : 3.10.2
-- CMake command : /usr/bin/cmake
-- System : Linux
-- C++ compiler : /usr/bin/c++
-- C++ compiler version : 7.5.0
-- CXX flags : -Wnon-virtual-dtor
even though it's already installed via pip install pybind11.
Any thoughts on how to move forward with this issue ?

Upgrade you protobuf version seems the issue is with default Ubuntu 18.04 repo protobuf package.
simmilar issue was reported on opencv forum.
github issue
Installation of protobuf 3.5.1 fixes this, maybe error in the original
install process. I installed protobuf 3.5.1 from this release
Install steps
./autogen.sh
./configure
make
make install

conda install and conda build result in different dependency versions

I'm trying to build a package which includes h5py. When using conda build, it seems to install the wrong version of the dependency. It installs 3.2.1-py37h6c542dc_0, which includes hdf5: 1.10.6-nompi_h6a2412b_1114.
The problem is that this hdf5 lib, seems to have these setting:
(Read-Only) S3 VFD: yes
This causes an error for me. When just running conda install h5py==3.2.1, it does install the right version (hdf5-1.10.6-nompi_h3c11f04_101).
Why is there a difference?

"Why is there a difference?
Using conda install h5py=3.2.1 additionally includes all the previous constraints in the current environment, whereas during a conda build run, a new environment is created only with requirements that the package specifies. That is, it is more like running conda create -n foo h5py=3.2.1.
So, that covers the mechanism, but we can also look at the particular package dependencies to see why the current environment constrains to the older hdf5-1.10.6-nompi_h3c11f04_101, which OP states is preferred. Here is the package info for the two:
hdf5-1.10.6-nompi_h6a2412b_1114
$ mamba search --info conda-forge/linux-64::hdf5[version='1.10.6',build='nompi_h6a2412b_1114']
hdf5 1.10.6 nompi_h6a2412b_1114
-------------------------------
file name : hdf5-1.10.6-nompi_h6a2412b_1114.tar.bz2
name : hdf5
version : 1.10.6
build : nompi_h6a2412b_1114
build number: 1114
size : 3.1 MB
license : LicenseRef-HDF5
subdir : linux-64
url : https://conda.anaconda.org/conda-forge/linux-64/hdf5-1.10.6-nompi_h6a2412b_1114.tar.bz2
md5 : 0a2984b78f51148d7ff6219abe73509e
timestamp : 2021-01-08 23:10:11 UTC
dependencies:
- libcurl >=7.71.1,<8.0a0
- libgcc-ng >=9.3.0
- libgfortran-ng
- libgfortran5 >=9.3.0
- libstdcxx-ng >=9.3.0
- openssl >=1.1.1i,<1.1.2a
- zlib >=1.2.11,<1.3.0a0
hdf5-1.10.6-nompi_h3c11f04_101
$ mamba search --info conda-forge/linux-64::hdf5[version='1.10.6',build='nompi_h3c11f04_101']
hdf5 1.10.6 nompi_h3c11f04_101
------------------------------
file name : hdf5-1.10.6-nompi_h3c11f04_101.tar.bz2
name : hdf5
version : 1.10.6
build : nompi_h3c11f04_101
build number: 101
size : 3.0 MB
license : HDF5
subdir : linux-64
url : https://conda.anaconda.org/conda-forge/linux-64/hdf5-1.10.6-nompi_h3c11f04_101.tar.bz2
md5 : 9f1ccc4d36edf8ea15ce19f52cf6d601
timestamp : 2020-07-31 12:26:29 UTC
dependencies:
- libgcc-ng >=7.5.0
- libgfortran-ng >=7,<8.0a0
- libstdcxx-ng >=7.5.0
- zlib >=1.2.11,<1.3.0a0
The difference here is that the latter works with older versions of libgcc-ng, libstdcxx-ng, and libgfortran-ng (below 9.3.0), as well as has no constraint on openssl or libcurl. So, we can guess that the current environment where the conda install h5py=3.2.1 was invoked has one of these restrictions.

Enable multi-threading on Caffe2

When compiling my program using Caffe2 I get this warnings:
[E init_intrinsics_check.cc:43] CPU feature avx is present on your machine, but the Caffe2 binary is not compiled with it. It means you may not get the full speed of your CPU.
[E init_intrinsics_check.cc:43] CPU feature avx2 is present on your machine, but the Caffe2 binary is not compiled with it. It means you may not get the full speed of your CPU.
[E init_intrinsics_check.cc:43] CPU feature fma is present on your machine, but the Caffe2 binary is not compiled with it. It means you may not get the full speed of your CPU.
Since I do want to get multi-threading support for Caffe2, I've searched what to do. I've found that Caffe2 has to be re-compiled setting some arguments while creating the cmake or in the CMakeLists.
Since I already had installed pytorch in a conda env, I have first uninstalled Caffe2 with:
pip uninstall -y caffe2
Then I've followed the instructions from the Caffe2 docs, to build it from sources.
I first installed the dependencies as indicated. Then I downloaded pytorch inside my conda env with:
git clone https://github.com/pytorch/pytorch.git && cd pytorch
git submodule update --init --recursive
At this time I think is the moment to change the pytorch\caffe2\CMakeLists file just downloaded. I have read that in order to enable the multi-threading support is sufficient to enable the option USE_NATIVE_ARCH inside this CMakeLists, however I'm not able to find such option where I'm looking. Maybe I'm doing something wrong. Any thoughts? Thanks.
Here some details about my platform:
I'm on macOS Big Sur
My python version is 3.8.5
UPDATE:
To answer Nega this is what I've got:
python3 -c 'import torch; print(torch.__config__.parallel_info())'
ATen/Parallel:
at::get_num_threads() : 1
at::get_num_interop_threads() : 4
OpenMP not found
Intel(R) Math Kernel Library Version 2020.0.2 Product Build 20200624 for Intel(R) 64 architecture applications
mkl_get_max_threads() : 4
Intel(R) MKL-DNN v0.21.1 (Git Hash 7d2fd500bc78936d1d648ca713b901012f470dbc)
std::thread::hardware_concurrency() : 8
Environment variables:
OMP_NUM_THREADS : [not set]
MKL_NUM_THREADS : [not set]
ATen parallel backend: OpenMP
UPDATE 2:
It turned out that the Clang that comes with XCode doesn't support OpenMP. The gcc that I was using was just a symlink to Clang. In fact after running gcc --version I got:
Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/include/c++/4.2.1
Apple clang version 12.0.0 (clang-1200.0.32.29)
Target: x86_64-apple-darwin20.3.0
Thread model: posix
InstalledDir: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin
I installed from Homebrew gcc-10 and set the alias like this alias gcc='gcc-10'. In fact now with gcc --version this is what I get:
gcc-10 (Homebrew GCC 10.2.0_4) 10.2.0
Copyright (C) 2020 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
I've also tried a simple Hello World for OpenMP using 8 threads and everything seems to be working. However after re-running the command:
python3 -c 'import torch; print(torch.__config__.parallel_info())'
I get the same outcome. Any thoughts?

AVX, AVX2, and FMA are CPU instruction sets and are not related to multi-threading. If the pip package for pytorch/caffe2 used these instructions on a CPU that didn't support them, the software wouldnt work. Pytorch, installed via pip comes with multi-threading enabled though. You can confirm this with torch.__config__.parallel_info()
❯ python3 -c 'import torch; print(torch.__config__.parallel_info())'
ATen/Parallel:
at::get_num_threads() : 6
at::get_num_interop_threads() : 6
OpenMP 201107 (a.k.a. OpenMP 3.1)
omp_get_max_threads() : 6
Intel(R) Math Kernel Library Version 2020.0.1 Product Build 20200208 for Intel(R) 64 architecture applications
mkl_get_max_threads() : 6
Intel(R) MKL-DNN v1.6.0 (Git Hash 5ef631a030a6f73131c77892041042805a06064f)
std::thread::hardware_concurrency() : 12
Environment variables:
OMP_NUM_THREADS : [not set]
MKL_NUM_THREADS : [not set]
ATen parallel backend: OpenMP
That being said, if you still want to continue building pytorch and caffe2 from source, the flag your looking for, USE_NATIVE is in pytorch/CMakeLists.txt, one level up from caffe2. Edit that file and change USE_NATIVE to ON. Then continue building pytorch with python3 setup.py build. Note that the flag USE_NATIVE doesn't do what you think it does. It only allows the building of MKL-DNN with CPU native optimization flags. It does not trickle down to caffe2 (except where caffe2 use MKL-DNN obviously.)

Cannot install nlopt python module

I am trying to install nlopt onto macOS 10.15.5. I downloaded the nlopt-2.6.2.tar.gz file from the NLopt docs and ran the following from the nlopt-2.6.2 directory:
mkdir build
cd build
cmake -DNLOPT_OCTAVE=Off -DNLOPT_MATLAB=Off -DNLOPT_GUILE=Off ..
make
sudo make install
I got the following output: cmake.txt.
The header file (nlopt.h) installs correctly to /usr/local/include and the dynamic library (libnlopt.dylib) installs correctly to /usr/local/lib/, but neither the dist-info file nor the nlopt module itself installs.
I have also tried installing via pip, brew, and conda, none of which have worked. I have also tried cloning from this Github, which didn't work either.
I appreciate any help with this, because I am completely lost. I am relatively new to this kind of stuff, and I couldn't find any good answers online.

The official docs are somewhat laconic about the exact steps required for building nlopt with Python bindings. First of all, you'll need SWIG installed:
$ brew install swig
Then, you'll need numpy to be available for the target Python interpreter. It is already preinstalled for the system Python, otherwise install it via Homebrew or pip, depending on your Python installation.
Now run cmake:
$ cmake -DNLOPT_GUILE=OFF -DNLOPT_MATLAB=OFF -DNLOPT_OCTAVE=OFF -DNLOPT_TESTS=OFF
This will build bindings against the default Python 2.7 installation preinstalled on MacOS. If you need to build against custom Python installation (e.g. when you've installed Python 3 via Homebrew or PKG installer from https://www.python.org/downloads), pass it via the PYTHON_EXECUTABLE arg:
$ cmake -DNLOPT_GUILE=OFF -DNLOPT_MATLAB=OFF -DNLOPT_OCTAVE=OFF -DNLOPT_TESTS=OFF -DPYTHON_EXECUTABLE=/usr/local/bin/python3
Inspect the log now - Python, SWIG and numpy headers should be successfully located. Example output snippet (you may have different paths/versions printed):
-- Found PythonInterp: /usr/local/bin/python3.8 (found version "3.8.3")
-- Found PythonLibs: /Library/Frameworks/Python.framework/Versions/3.8/lib/libpython3.8.dylib (found suitable exact version "3.8.3")
-- Found NumPy: /Users/hoefling/Library/Python/3.8/lib/python/site-packages/numpy/core/include (found version "1.19")
-- Found SWIG: /usr/local/bin/swig (found version "4.0.2")
If any of those conditions is not satisfied (e.g. you see Could NOT find NumPy, Could NOT find PythonLibs or Could NOT find SWIG), then stop and make sure the configuration succeeds before proceeding next.
Now compile:
$ make
...
Scanning dependencies of target nlopt_python_swig_compilation
[ 96%] Swig compile nlopt.i for python
[ 96%] Built target nlopt_python_swig_compilation
Scanning dependencies of target nlopt_python
[ 98%] Building CXX object src/swig/CMakeFiles/nlopt_python.dir/CMakeFiles/nlopt_python.dir/nloptPYTHON_wrap.cxx.o
[100%] Linking CXX shared module _nlopt.so
[100%] Built target nlopt_python
Install:
$ make install
...
-- Installing: /usr/local/lib/python3.8/site-packages/nlopt.py
-- Installing: /usr/local/lib/python3.8/site-packages/_nlopt.so
Test the Python bindings are importable:
$ python -c "import nlopt; print(nlopt.__version__)"
2.6.2

Cmake is not able to find Python-libraries

Getting this error:
sudo: unable to resolve host coderw#ll
-- Could NOT find PythonLibs (missing: PYTHON_LIBRARIES PYTHON_INCLUDE_DIRS)
CMake Error at /usr/share/cmake-2.8/Modules/FindPackageHandleStandardArgs.cmake:108
(message):
Could NOT find PythonInterp (missing: PYTHON_EXECUTABLE)
Call Stack (most recent call first):
/usr/share/cmake-2.8/Modules/FindPackageHandleStandardArgs.cmake:315
(_FPHSA_FAILURE_MESSAGE)
/usr/share/cmake-2.8/Modules/FindPythonInterp.cmake:139
(FIND_PACKAGE_HANDLE_STANDARD_ARGS)
Code/cmake/Modules/FindNumPy.cmake:10 (find_package)
CMakeLists.txt:114 (find_package)
-- Configuring incomplete, errors occurred!
See also "/home/coderwall/Desktop/rdkit/build/CMakeFiles/CMakeOutput.log".
See also "/home/coderwall/Desktop/rdkit/build/CMakeFiles/CMakeError.log".
I have already installed:
sudo apt-get install python-dev
Environment variable are already set as follow:
PYTHON_INCLUDE_DIRS=/usr/include/python2.7
PYTHON_LIBRARIES=/usr/lib/python2.7/config/libpython2.7.so
Location of python.h : /usr/lib/include/python2.7/python.h
Location of python libs: /usr/lib/python2.7/
How to solve this?

You can fix the errors by appending to the cmake command the -DPYTHON_LIBRARY and -DPYTHON_INCLUDE_DIR flags filled with the respective folders.
Thus, the trick is to fill those parameters with the returned information from the python interpreter, which is the most reliable. This may work independently of your python location/version (also for Anaconda users):
$ cmake .. \
-DPYTHON_INCLUDE_DIR=$(python -c "import sysconfig; print(sysconfig.get_path('include'))") \
-DPYTHON_LIBRARY=$(python -c "import sysconfig; print(sysconfig.get_config_var('LIBDIR'))")
If the version of python that you want to link against cmake is Python3.X and the default python symlink points to Python2.X, python3 -c ... can be used instead of python -c ....
In case that the error persists, you may need to update the cmake to a higher version as stated by #pdpcosta and repeat the process again.

For me this is helpful:
# if using python2
apt-get install python-dev
# if using python3
apt-get install python3-dev

I hit the same issue,and discovered the error message gives misleading variable names. Try setting the following (singular instead of plural):
PYTHON_INCLUDE_DIR=/usr/include/python2.7
PYTHON_LIBRARY=/usr/lib/python2.7/config/libpython2.7.so
The (plural) variables you see error messages about are values that the PythonLibs sets up when it is initialised correctly.

Even after adding -DPYTHON_INCLUDE_DIR and -DPYTHON_LIBRARY as suggested above, I was still facing the error Could NOT find PythonInterp. What solved it was adding -DPYTHON_EXECUTABLE:FILEPATH= to cmake as suggested in https://github.com/pybind/pybind11/issues/99#issuecomment-182071479:
cmake .. \
-DPYTHON_INCLUDE_DIR=$(python -c "from distutils.sysconfig import get_python_inc; print(get_python_inc())") \
-DPYTHON_LIBRARY=$(python -c "import distutils.sysconfig as sysconfig; print(sysconfig.get_config_var('LIBDIR'))") \
-DPYTHON_EXECUTABLE:FILEPATH=`which python`

I was facing this problem while trying to compile OpenCV 3 on a Xubuntu 14.04 Thrusty Tahr system.
With all the dev packages of Python installed, the configuration process was always returning the message:
Could NOT found PythonInterp: /usr/bin/python2.7 (found suitable version "2.7.6", minimum required is "2.7")
Could NOT find PythonLibs (missing: PYTHON_INCLUDE_DIRS) (found suitable exact version "2.7.6")
Found PythonInterp: /usr/bin/python3.4 (found suitable version "3.4", minimum required is "3.4")
Could NOT find PythonLibs (missing: PYTHON_LIBRARIES) (Required is exact version "3.4.0")
The CMake version available on Thrusty Tahr repositories is 2.8.
Some posts inspired me to upgrade CMake.
I've added a PPA CMake repository which installs CMake version 3.2.
After the upgrade everything ran smoothly and the compilation was successful.

Some last version of Ubuntu installs Python 3.4 by default and the CMake version from Ubuntu (2.8) only searches up to Python 3.3.
Try to add set(Python_ADDITIONAL_VERSIONS 3.4) before the find_package statement.
Remember to clean CMakeCache.txt too.

This problem can also happen in Windows. Cmake looks into the registry and sometimes python values are not set. For those with similar problem:
http://ericsilva.org/2012/10/11/restoring-your-python-registry-in-windows/
Just create a .reg file to set the necessary keys and edit accordingly to match your setup.
Windows Registry Editor Version 5.00
[HKEY_CURRENT_USER\Software\Python]
[HKEY_CURRENT_USER\Software\Python\Pythoncore]
[HKEY_CURRENT_USER\Software\Python\Pythoncore\2.6]
[HKEY_CURRENT_USER\Software\Python\Pythoncore\2.6\InstallPath]
#="C:\\python26"
[HKEY_CURRENT_USER\Software\Python\Pythoncore\2.6\PythonPath]
#="C:\\python26;C:\\python26\\Lib\\;C:\\python26\\DLLs\\"
[HKEY_CURRENT_USER\Software\Python\Pythoncore\2.7]
[HKEY_CURRENT_USER\Software\Python\Pythoncore\2.7\InstallPath]
#="C:\\python27"
[HKEY_CURRENT_USER\Software\Python\Pythoncore\2.7\PythonPath]
#="C:\\python27;C:\\python27\\Lib\\;C:\\python27\\DLLs\\"

Note that if you are using cMake version 3.12 or later, variable PythonInterp and PythonLibs has been changed into Python.
So we use:
find_package(Python ${PY_VERSION} REQUIRED)
instead of:
find_package(PythonInterp ${PY_VERSION} REQUIRED)
find_package(PythonLibs ${PY_VERSION} REQUIRED)
see https://cmake.org/cmake/help/v3.12/module/FindPython.html for details.

I had upgraded to python3.8 on my system and had an incomplete install. Managed to fix it by installing the rest of the 3.8 packages:
sudo apt-get install python3.8 python3.8-dev python3.8-distutils python3.8-venv

Paste this into your CMakeLists.txt:
# find python
execute_process(COMMAND python-config --prefix OUTPUT_VARIABLE PYTHON_SEARCH_PATH)
string(REGEX REPLACE "\n$" "" PYTHON_SEARCH_PATH "${PYTHON_SEARCH_PATH}")
file(GLOB_RECURSE PYTHON_DY_LIBS ${PYTHON_SEARCH_PATH}/lib/libpython*.dylib ${PYTHON_SEARCH_PATH}/lib/libpython*.so)
if (PYTHON_DY_LIBS)
list(GET PYTHON_DY_LIBS 0 PYTHON_LIBRARY)
message("-- Find shared libpython: ${PYTHON_LIBRARY}")
else()
message(WARNING "Cannot find shared libpython, try find_package")
endif()
find_package(PythonInterp)
find_package(PythonLibs ${PYTHON_VERSION_STRING} EXACT)

In case that might help, I found a workaround for a similar problem, looking at the cmake doc :
https://cmake.org/cmake/help/v3.0/module/FindPythonLibs.html
You must set two env vars for cmake to find coherent versions.
Unfortunately this is not a generic solution...
cmake -DPYTHON_LIBRARY=${HOME}/.pyenv/versions/3.8.0/lib/libpython3.8.a -DPYTHON_INCLUDE_DIR=${HOME}/.pyenv/versions/3.8.0/include/python3.8/ cern_root/

In Python 3.2 and onward distutils.sysconfig is deprecated in favor of sysconfig.
To get all the variable names in data structure and inspect the situation we can use get_paths function
import sysconfig
sysconfig.get_paths()
which will return us a dict with all the relevant variable names as keys, and corresponding paths as values.
When we know the key we can get the value dynamically,
>>> sysconfig.get_path("include")
'C:\\Program Files\\Python310\\Include'
However, the most convenient feature of sysutils for this situation is that it can list all the variables at once from the command-prompt by invoking python -m sysconfig:
$ python -m sysconfig
Platform: "macosx-10.4-i386"
Python version: "3.2"
Current installation scheme: "posix_prefix"
Paths:
data = "/usr/local"
include = "/Users/tarek/Dev/svn.python.org/py3k/Include"
platinclude = "."
platlib = "/usr/local/lib/python3.2/site-packages"
platstdlib = "/usr/local/lib/python3.2"
purelib = "/usr/local/lib/python3.2/site-packages"
scripts = "/usr/local/bin"
stdlib = "/usr/local/lib/python3.2"
Variables:
AC_APPLE_UNIVERSAL_BUILD = "0"
AIX_GENUINE_CPLUSPLUS = "0"
AR = "ar"
ARFLAGS = "rc"
...

Maybe below command line works for you if all of above methods didn't work.
sudo apt-get install cython cython3

On Ubuntu 20 at least, since Python 2.7 is deprecated, you can:
sudo apt install python2.7-dev
And then PYTHON_INCLUDE_DIR can be set to /usr/include/python2.7