Skip to content

Latest commit

 

History

History
456 lines (334 loc) · 19.7 KB

install.rst

File metadata and controls

456 lines (334 loc) · 19.7 KB

Installation

Using conda

Conda is a cross platform package manager and provides the easiest mechanism to install cyipopt on Linux, Mac, and Windows. Once conda is installed, install cyipopt from the Conda Forge channel with:

$ conda install -c conda-forge cyipopt

The above command will install binary versions of all the necessary dependencies as well as cyipopt. Conda Forge supplies a basic build of Ipopt that is suitable for many use cases. You will have to install from source if you want a customized Ipopt installation.

From source

To begin installing from source you will need to install the following dependencies:

  • C/C++ compiler
  • pkg-config [only for Linux and Mac]
  • Ipopt >=3.12 [>= 3.13 on Windows]
  • Python 3.9+
  • setuptools >=44.1.1
  • cython >=0.29.28
  • NumPy >=1.25
  • SciPy >=1.8 [optional]

The binaries and header files of the Ipopt package can be obtained from http://www.coin-or.org/download/binary/Ipopt/. These include a version compiled against the MKL library. Or you can build Ipopt from source. The remaining dependencies can be installed with conda or other package managers.

On Linux and Mac

For Linux and Mac, the ipopt executable should be in your path and discoverable by pkg-config, i.e. this command should return a valid result:

$ pkg-config --libs --cflags ipopt

You will need to install Ipopt in a system location or set LD_LIBRARY_PATH if pkg-config does not find the executable.

Once all the dependencies are installed, execute:

$ python setup.py install

to build and install the package.

From source on Windows

Install the dependencies with conda (Anaconda or Miniconda):

$ conda.exe install -c conda-forge numpy cython setuptools

Or alternatively with pip:

$ pip install numpy cython setuptools

Additionally, make sure you have a C compiler setup to compile Python C extensions, e.g. Visual C++. Build tools for VS2019 https://visualstudio.microsoft.com/downloads/#build-tools-for-visual-studio-2019 have been tested to work for conda Python 3.7 (see #52).

Download and extract the cyipopt source code from Github or PyPi.

Obtain IPOPT one of two ways:

  1. Using official IPOPTs binaries:

Download the latest precompiled version of Ipopt that includes the DLL files from https://github.com/coin-or/Ipopt/releases. Note that the current setup only supports Ipopt >= 3.13.0. The build 3.13.3 of Ipopt has been confirmed to work and can be downloaded from Ipopt-3.13.3-win64-msvs2019-md.zip. After Ipopt is extracted, the bin, lib and include folders should be in the root cyipopt directory, i.e. adjacent to the setup.py file. Alternatively, you can set the environment variable IPOPTWINDIR to point to the Ipopt directory that contains the bin, lib and include directories.

  1. Using Conda Forge's IPOPT binary:

If using conda, you can install an IPOPT binary from Conda Forge:

$ conda.exe install -c conda-forge ipopt

The environment variable IPOPTWINDIR should then be set to USECONDAFORGEIPOPT.

Finally, execute:

$ python setup.py install

NOTE: It is advised to use the Anaconda or Miniconda distributions and not the official python.org distribution. Even though it has been tested to work with the latest builds, it is well-known for causing issues. (see #52).

On Ubuntu 22.04 Using APT Dependencies

All of the dependencies can be installed with Ubuntu's package manager:

$ apt install build-essential pkg-config python3-pip python3-dev cython3 python3-numpy coinor-libipopt1v5 coinor-libipopt-dev

You can then install cyipopt from the PyPi release with:

$ python3 -m pip install cyipopt

Or you use a local copy with:

$ cd /cyipopt/source/directory/
$ python3 setup.py install

On Ubuntu 18.04 Using APT Dependencies

All of the dependencies can be installed with Ubuntu's package manager:

$ sudo apt install build-essential pkg-config python-dev cython python-numpy coinor-libipopt1v5 coinor-libipopt-dev

The NumPy and IPOPT libs and headers are installed in standard locations, so you should not need to set LD_LIBRARY_PATH or PKG_CONFIG_PATH.

Now run python setup.py build to compile cyipopt. In the output of this command you should see two calls to gcc for compiling and linking. Make sure both of these are pointing to the correct libraries and headers. They will look something like this (formatted and commented for easy viewing here):

$ python setup.py build
...
x86_64-linux-gnu-gcc -pthread -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-prototypes -fno-strict-aliasing
  -Wdate-time -D_FORTIFY_SOURCE=2 -g -fdebug-prefix-map=/build/python2.7-3hk45v/python2.7-2.7.15~rc1=.
  -fstack-protector-strong -Wformat -Werror=format-security -fPIC
  -I/usr/local/include/coin  # points to IPOPT headers
  -I/usr/local/include/coin/ThirdParty  # points to IPOPT third party headers
  -I/usr/lib/python2.7/dist-packages/numpy/core/include  # points to NumPy headers
  -I/usr/include/python2.7  # points to Python 2.7 headers
  -c src/cyipopt.c -o build/temp.linux-x86_64-2.7/src/cyipopt.o
x86_64-linux-gnu-gcc -pthread -shared -Wl,-O1 -Wl,-Bsymbolic-functions -Wl,-Bsymbolic-functions -Wl,-z,relro
  -fno-strict-aliasing -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-prototypes -Wdate-time -D_FORTIFY_SOURCE=2 -g
  -fdebug-prefix-map=/build/python2.7-3hk45v/python2.7-2.7.15~rc1=. -fstack-protector-strong -Wformat
  -Werror=format-security -Wl,-Bsymbolic-functions -Wl,-z,relro -Wdate-time -D_FORTIFY_SOURCE=2 -g
  -fdebug-prefix-map=/build/python2.7-3hk45v/python2.7-2.7.15~rc1=. -fstack-protector-strong -Wformat
  -Werror=format-security build/temp.linux-x86_64-2.7/src/cyipopt.o
  -L/usr/local/lib
  -L/lib/../lib
  -L/usr/lib/../lib
  -L/usr/lib/gcc/x86_64-linux-gnu/5
  -L/usr/lib/gcc/x86_64-linux-gnu/5/../../..
  -L/usr/lib/gcc/x86_64-linux-gnu/5/../../../../lib
  -L/usr/lib/gcc/x86_64-linux-gnu/5/../../../x86_64-linux-gnu
  -lipopt -llapack -lblas -lm -ldl -lcoinmumps -lblas -lgfortran -lm -lquadmath  # linking to relevant libs
  -lcoinhsl -llapack -lblas -lgfortran -lm -lquadmath -lcoinmetis  # linking to relevant libs
  -o build/lib.linux-x86_64-2.7/cyipopt.so
...

You can check that everything linked correctly with ldd:

$ ldd build/lib.linux-x86_64-2.7/cyipopt.so
linux-vdso.so.1 (0x00007ffc1677c000)
libipopt.so.0 => /usr/local/lib/libipopt.so.0 (0x00007fcdc8668000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fcdc8277000)
libcoinmumps.so.0 => /usr/local/lib/libcoinmumps.so.0 (0x00007fcdc7eef000)
libcoinhsl.so.0 => /usr/local/lib/libcoinhsl.so.0 (0x00007fcdc7bb4000)
liblapack.so.3 => /usr/lib/x86_64-linux-gnu/liblapack.so.3 (0x00007fcdc732e000)
libblas.so.3 => /usr/lib/x86_64-linux-gnu/libblas.so.3 (0x00007fcdc70d3000)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fcdc6ecf000)
libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007fcdc6b46000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fcdc67a8000)
/lib64/ld-linux-x86-64.so.2 (0x00007fcdc8d20000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fcdc6590000)
libcoinmetis.so.0 => /usr/local/lib/libcoinmetis.so.0 (0x00007fcdc6340000)
libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fcdc600f000)
libopenblas.so.0 => /usr/lib/x86_64-linux-gnu/libopenblas.so.0 (0x00007fcdc3d69000)
libgfortran.so.4 => /usr/lib/x86_64-linux-gnu/libgfortran.so.4 (0x00007fcdc398a000)
libquadmath.so.0 => /usr/lib/x86_64-linux-gnu/libquadmath.so.0 (0x00007fcdc374a000)
libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007fcdc352b000)

And finally install the package into Python's default package directory:

$ python setup.py install

Note that you may or may not want to install this package system wide, i.e. prepend sudo to the above command, but it is safest to install into your user space, i.e. what pip install --user does, or setup a virtual environment with tools like venv or conda. If you use virtual environments you will need to be careful about selecting headers and libraries for packages in or out of the virtual environments in the build step. Note that cython, and numpy could alternatively be installed using Python specific package managers, e.g. pip install cython numpy.

On Ubuntu 18.04 with Custom Compiled IPOPT

Install system wide dependencies:

$ sudo apt install pkg-config python-dev wget
$ sudo apt build-dep coinor-libipopt1v5

Install pip so all Python packages can be installed via pip:

$ sudo apt install python-pip

Then use pip to install the following packages:

$ pip install --user numpy cython

Compile Ipopt

The Ipopt compilation instructions are derived from https://www.coin-or.org/Ipopt/documentation/node14.html. If you get errors, start there for help.

Download Ipopt source code. Choose the version that you would like to have from <https://www.coin-or.org/download/source/Ipopt/>. For example:

$ cd ~
$ wget https://www.coin-or.org/download/source/Ipopt/Ipopt-3.12.11.tgz

Extract the Ipopt source code:

$ tar -xvf Ipopt-3.12.11.tgz

Create a temporary environment variable pointing to the Ipopt directory:

$ export IPOPTDIR=~/Ipopt-3.12.11

To use linear solvers other than the default mumps, e.g. ma27, ma57, ma86 solvers, the HSL package are needed. HSL can be downloaded from its official website <http://www.hsl.rl.ac.uk/ipopt/>.

Extract HSL source code after you get it. Rename the extracted folder to coinhsl and copy it in the HSL folder: Ipopt-3.12.11/ThirdParty/HSL

Build Ipopt:

$ mkdir $IPOPTDIR/build
$ cd $IPOPTDIR/build
$ ../configure
$ make
$ make test

Add make install if you want a system wide install.

Set environment variables:

$ export IPOPT_PATH="~/Ipopt-3.12.11/build"
$ export PKG_CONFIG_PATH=$PKG_CONFIG_PATH:$IPOPT_PATH/lib/pkgconfig
$ export PATH=$PATH:$IPOPT_PATH/bin

Get help from this web-page if you get errors in setting environments:

https://stackoverflow.com/questions/13428910/how-to-set-the-environmental-variable-ld-library-path-in-linux

Now compile cyipopt. Download the cyipopt source code from PyPi, for example:

$ cd ~
$ wget https://files.pythonhosted.org/packages/05/57/a7c5a86a8f899c5c109f30b8cdb278b64c43bd2ea04172cbfed721a98fac/ipopt-0.1.9.tar.gz
$ tar -xvf ipopt-0.1.8.tar.gz
$ cd ipopt

Compile cyipopt:

$ python setup.py build

If there is no error, then you have compiled cyipopt successfully

Check that everything linked correctly with ldd

$ ldd build/lib.linux-x86_64-2.7/cyipopt.so
linux-vdso.so.1 (0x00007ffe895e1000)
libipopt.so.1 => /home/<username>/Ipopt-3.12.11/build/lib/libipopt.so.1 (0x00007f74efc2a000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f74ef839000)
libcoinmumps.so.1 => /home/<username>/Ipopt-3.12.11/build/lib/libcoinmumps.so.1 (0x00007f74ef4ae000)
libcoinhsl.so.1 => /home/<username>/Ipopt-3.12.11/build/lib/libcoinhsl.so.1 (0x00007f74ef169000)
liblapack.so.3 => /usr/lib/x86_64-linux-gnu/liblapack.so.3 (0x00007f74ee8cb000)
libblas.so.3 => /usr/lib/x86_64-linux-gnu/libblas.so.3 (0x00007f74ee65e000)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f74ee45a000)
libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007f74ee0d1000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f74edd33000)
/lib64/ld-linux-x86-64.so.2 (0x00007f74f02c0000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f74edb1b000)
libcoinmetis.so.1 => /home/<username>/Ipopt-3.12.11/build/lib/libcoinmetis.so.1 (0x00007f74ed8ca000)
libgfortran.so.4 => /usr/lib/x86_64-linux-gnu/libgfortran.so.4 (0x00007f74ed4eb000)

Install cyipopt (prepend sudo if you want a system wide install):

$ python setup.py install

To use cyipopt you will need to set the LD_LIBRARY_PATH to point to your Ipopt install if you did not install it to a standard location. For example:

$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:~/Ipopt-3.12.11/build/lib

You can add this to your shell's configuration file if you want it set every time you open your shell, for example the following line can it can be added to your ~/.bashrc

$ echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/Ipopt-3.12.11/build/lib' >> ~/.bashrc

Now you should be able to run a cyipopt example:

$ cd test
$ python -c "import cyipopt"
$ python examplehs071.py

If it could be run successfully, the optimization will start with the following descriptions:

******************************************************************************
This program contains Ipopt, a library for large-scale nonlinear optimization.
 Ipopt is released as open source code under the Eclipse Public License (EPL).
         For more information visit http://projects.coin-or.org/Ipopt
******************************************************************************

This is Ipopt version 3.12.11, running with linear solver ma27.
...

Conda Forge binaries with HSL

On Linux

It is possible to use the HSL linear solvers with cyipopt installed via Conda Forge. To do so, first download the HSL source code tarball. The following explanation uses coinhsl-2014.01.10.tar.gz with conda installed on Ubuntu 20.04.

Create a conda environment with at least gfortran and cyipopt:

$ conda create -n hsl-test -c conda-forge gfortran cyipopt
$ conda activate hsl-test

You should now have an environment that includes ipopt. You can checked what ipopt is linked against like so:

(hsl-test) $ ldd ~/miniconda/envs/hsl-test/lib/libipopt.so
   linux-vdso.so.1 (0x00007ffcaf45b000)
   librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007f8965748000)
   liblapack.so.3 => /home/<username>/miniconda/envs/hsl-test/lib/./liblapack.so.3 (0x00007f89635fe000)
   libdmumps_seq-5.2.1.so => /home/<username>/miniconda/envs/hsl-test/lib/./libdmumps_seq-5.2.1.so (0x00007f89633d8000)
   libmumps_common_seq-5.2.1.so => /home/<username>/miniconda/envs/hsl-test/lib/./libmumps_common_seq-5.2.1.so (0x00007f8963377000)
   libpord_seq-5.2.1.so => /home/<username>/miniconda/envs/hsl-test/lib/./libpord_seq-5.2.1.so (0x00007f896335e000)
   libmpiseq_seq-5.2.1.so => /home/<username>/miniconda/envs/hsl-test/lib/./libmpiseq_seq-5.2.1.so (0x00007f8963352000)
   libesmumps-6.so => /home/<username>/miniconda/envs/hsl-test/lib/./libesmumps-6.so (0x00007f8963349000)
   libscotch-6.so => /home/<username>/miniconda/envs/hsl-test/lib/./libscotch-6.so (0x00007f89632b1000)
   libscotcherr-6.so => /home/<username>/miniconda/envs/hsl-test/lib/./libscotcherr-6.so (0x00007f89632ac000)
   libmetis.so => /home/<username>/miniconda/envs/hsl-test/lib/./libmetis.so (0x00007f8963237000)
   libgfortran.so.5 => /home/<username>/miniconda/envs/hsl-test/lib/./libgfortran.so.5 (0x00007f896308e000)
   libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f8963088000)
   libstdc++.so.6 => /home/<username>/miniconda/envs/hsl-test/lib/./libstdc++.so.6 (0x00007f8962edb000)
   libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f8962d8c000)
   libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f8962b9a000)
   /lib64/ld-linux-x86-64.so.2 (0x00007f8965a02000)
   libgcc_s.so.1 => /home/<username>/miniconda/envs/hsl-test/lib/./libgcc_s.so.1 (0x00007f8962b85000)
   libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f8962b62000)
   libgomp.so.1 => /home/<username>/miniconda/envs/hsl-test/lib/././libgomp.so.1 (0x00007f8962b2a000)
   libz.so.1 => /home/<username>/miniconda/envs/hsl-test/lib/././libz.so.1 (0x00007f8962b10000)
   libquadmath.so.0 => /home/<username>/miniconda/envs/hsl-test/lib/././libquadmath.so.0 (0x00007f8962ad6000)

Now navigate into the extracted HSL directory and configure HSL:

(hsl-test) $ cd /path/to/coinhsl-2014.01.10/
(hsl-test) $ ./configure \
   --prefix=/home/<username>/miniconda/envs/hsl-test/ \
   --with-blas="-L/home/<username>/miniconda/envs/hsl-test/lib/ -lblas" \
   LIBS="-llapack" \
   FC=/home/<username>/miniconda/envs/hsl-test/bin/gfortran \
   CC=/home/<username>/miniconda/envs/hsl-test/bin/gcc \

This tells HSL to install into your environment, link against the environment's blas and lapack libraries and to use the environment's gfortran and gcc compilers to build HSL. After configuring, build and install with:

(hsl-test) $ make
(hsl-test) $ make install

You should now find a shared HSL library in your environment. Check to make sure it is properly linked (especially blas):

(hsl-test) $ ldd ~/miniconda/envs/hsl-test/lib/libcoinhsl.so
   linux-vdso.so.1 (0x00007ffe2085a000)
   libopenblas.so.0 => /home/<username>/miniconda/envs/hsl-test/lib/libopenblas.so.0 (0x00007f72a1766000)
   libgfortran.so.5 => /home/<username>/miniconda/envs/hsl-test/lib/libgfortran.so.5 (0x00007f72a15bd000)
   libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f72a143f000)
   libgcc_s.so.1 => /home/<username>/miniconda/envs/hsl-test/lib/libgcc_s.so.1 (0x00007f72a142a000)
   libquadmath.so.0 => /home/<username>/miniconda/envs/hsl-test/lib/libquadmath.so.0 (0x00007f72a13f0000)
   libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f72a11fe000)
   libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f72a11d9000)
   /lib64/ld-linux-x86-64.so.2 (0x00007f72a39d4000)

Now, in your cyipopt script set the following two options:

problem.add_option('linear_solver', 'ma57')
problem.add_option('hsllib', 'libcoinhsl.so')

The various HSL solvers can be set with linear_solver and the hsllib name must be specified because the default name ipopt looks for is libhsl.so. Identify the shared library installed on your system and make sure the name provided for the hsllib option matches. For example, on macOS you may need problem.add_option('hsllib', 'libcoinhsl.dylib').

On Windows

On windows it could be possible to use HSL solvers with Conda version of cyipopt, first install cyipopt via standard call:

$ conda create -n hsl-test -c conda-forge cyipopt
$ conda activate hsl-test

Download HSL linear solvers, e.g. ma27, ma57, ma86 from its official website <http://www.hsl.rl.ac.uk/ipopt/>. Download the windows binaries option, in this example, we are using "CoinHSL Archive 2023.11.17 (windows binaries)" option. This will download a zipped file that contains a folder bin. Copy all the DLL files from that folder into your conda env\Library\bin folder (This folder should also contain the an ipopt dll file installed with cyipopt (in our case it was called ipopt-3.dll):

<conda_location>\envs\<env_name>\Library\bin

Once the DLL files are placed you should be able to access the HSL solvers using solver options:

problem.add_option('linear_solver', 'ma57')

If all works well, you should see the following when setting tee=True:

******************************************************************************
This program contains Ipopt, a library for large-scale nonlinear optimization.
Ipopt is released as open source code under the Eclipse Public License (EPL).
         For more information visit https://github.com/coin-or/Ipopt
******************************************************************************

This is Ipopt version 3.14.16, running with linear solver ma27.

This was tested on Windows 11 x64 with Ipopt version 3.14.16.