Skip to content

Latest commit

 

History

History
109 lines (82 loc) · 5.53 KB

orca_install.md

File metadata and controls

109 lines (82 loc) · 5.53 KB
Raw

ORCA detailed build and troubleshooting

This a detailed install guide. If you have not done it yet, please try the Quick Install steps before continuing. After this, if you are here, that means that there have been some errors when building the mex files.

Building mex files in GNU/Linux

make.m scripts could fail if the compiler is not correctly installed and configured. In those cases, please try mex -setup to choose a suitable compiler for mex. Make sure your compiler is accessible. Then type make to start the installation.

If neither make.m nor mex -setup works, the simplest way to compile all the algorithms is to use the Makefile included in ORCA. This will compile all the algorithms and clean intermediate object files.

Building mex files for Matlab from the GNU/Linux terminal

For building the mex files in MATLAB, you need to properly configure the MATLABDIR variable of Makefile, in order to point out to the MATLAB installation directory (for instance MATLABDIR = /usr/local/MATLAB/R2017b). Then, from the bash terminal:

$ cd src/
$ make

Building mex files for Octave from the GNU/Linux terminal

For building the mex files in Octave, you will need to configure the OCTAVEDIR variable in the Makefile. This variable has to point out to the Octave heather files (for instance, OCTAVEDIR = /usr/include/octave-4.0.0/octave/). Then, from the bash terminal:

$ cd src/
$ make octave

Building mex files in Windows

In Windows, we recommend compiling mex files from Octave/MATLAB console.

Building mex files in Windows for Octave

Default Octave installation provides mex command pre-configured with MinGW.

  1. Inside Octave's console, run make in folder src\Algorithms
  2. From src run runtests_single to check the installation.

Building mex files in Windows for Matlab

  1. Install a supported compiler. The easier way is to use the "Add-ons" assistant to download and install MinGW.
  2. Test basic C example to ensure mex is properly working.
  3. From the MATLAB's console, run make in src\Algorithms.
  4. Then run runtests_single in src to check the instalation.

We provide binaries and dlls for 'ORBoost', because building this method in Windows can be very complex. Make will unpack all the binary files. If you need to compile your own binaries, these are the steps:

  1. Install w64-mingw32.
  2. Open a terminal by pressing Windows icon and type cmd.exe.
  3. Set Windows path to your w64-mingw32 installation binaries dir, for instance:
set PATH=C:\Program Files\mingw-w64\x86_64-7.2.0-posix-seh-rt_v5-rev0\mingw64\bin;"%PATH%"
  1. Move to directory orca\src\Algorithms\orensemble\orensemble.
  2. Run mingw32-make.exe Makefile.win all.

Individually compiling the algorithms

If you are not able to compile all the algorithms using the above methods, we recommend individually compiling them. Under GNU/Linux, please, edit the files src/Algorithms/libsvm-rank-2.81/matlab/Makefile, src/Algorithms/libsvm-weights-3.12/matlab/Makefile, src/Algorithms/SVOREX/Makefile and src/Algorithms/SVORIM/Makefile. Make sure that the variables MATLABDIR or OCTAVEDIR are correctly pointing to the folders. For MATLAB, you can also make a symbolic link to your current Matlab installation folder:

$ sudo ln -s /path/to/matlab /usr/local/matlab

The following subsections provides individual instructions for compiling each of the dependencies in case the global Makefile still fails or for those which are working in other operating systems.

libsvm-weights-3.12

These instructions are adapted from the corresponding README of libsvm. First, you should open MATLAB/Octave console and then cd to the directory src/Algorithms/libsvm-weights-3.12/matlab. After that, try to compile the MEX files using make.m (from the MATLAB/Octave console):

>> cd src/Algorithms/libsvm-weights-3.12/matlab
>> make

libsvm-rank-2.81

To compile this dependency, the instructions are similar to those of libsvm-weights-3.12 (from the MATLAB/Octave console):

>> cd src/Algorithms/libsvm-rank-2.81/matlab
>> make

SVOREX and SVORIM

For both algorithms, please use the make.m file included in them (from the MATLAB/Octave console):

>> cd src/Algorithms/SVOREX
>> make
>> cd ..
>> cd SVORIM
>> make

orensemble

We have not prepared a proper MEX interface for ORBoost, so the binary files of this algorithm should be compiled and are then invoked directly from Matlab. For compiling the ORBoost algorithm, you should uncompress the file orsemble.tar.gz and compile the corresponding source code. In GNU/Linux, this can be done by (from the bash console):

$ cd src/Algorithms/orensemble
$ tar zxf orensemble.tar.gz
$ cd orensemble/
$ make
g++ -Ilemga-20060516/lemga -Wall -Wshadow -Wcast-qual -Wpointer-arith -Wconversion -Wredundant-decls -Wwrite-strings -Woverloaded-virtual -D NDEBUG -O3 -funroll-loops -c -o robject.o lemga-20060516/lemga/object.cpp
...

Then, you should move the binary files to .. folder and clean the folder (from the bash console):

$ mv boostrank-predict ../
$ mv boostrank-train ../
$ cd ..
$ rm -Rf orensemble