A simple python module containing an in-place linear rank filter optimized in C++.
This package is designed to provide an efficient linear rank order filter written in C++ with Python bindings. It can take single or double precision floats as input. It was needed as the equivalent percentile filter in SciPy was found to be too slow and unnecessarily general. There was no equivalent in VIGRA.
In order to build this package, the following requirements are needed.
- Python (2.7.x or 3.5.x)
- Boost (1.56.0 or later)
- NumPy (1.7.0 or later)
- Cython (0.23.0 or later)
- Setuptools (18.0 or later)
- VIGRA, optional for testing (1.11.0 or later)
The easiest way to install is to install our conda package. Alternatively, one can install from pip, but this will require a C++ compiler and a recent version of setuptools.
There are several ways to build the package.
- Standard Python build and install.
- Conda recipe build and install.
- CMake build and install.
The vanilla install in any of these forms should be basically equivalent.
To start simply clone the repo and change directory to the repo.
git clone https://github.com/nanshe-org/rank_filter cd rank_filter
To build/install with Python directly, simply run the following command.
python setup.py install
To build/install with Conda, simply run the following command.
conda build rank_filter.recipe conda install --use-local rank_filter.
In order to find Boost includes and libraries, the directory Boost was installed to must be set as BOOST_ROOT. Similarly, in order to find VIGRA includes and libraries, the directory VIGRA was installed to must be set as VIGRA_ROOT. CMake will also pick them up as options.
cmake -DBOOST_ROOT=<path-to-Boost-root> -DVIGRA_ROOT=<path-to-VIGRA-root> .
Also the CMake installer will also pick these variables up if they are set in the environment and not provided.
export BOOST_ROOT=<path-to-Boost-root> export VIGRA_ROOT=<path-to-VIGRA-root> cmake .
Additionally, the preferred python interpreter can be set by using the PYTHON_EXECUTABLE variable.
Before building the Python bindings it is worth checking if the C++ code passes its own test suite. This can be done using make with the command below. It is not required to run this stage, but it will be run every time when building. These test are no guarantee that the Python module will work. All they verify is that the C++ code works.
Building is done easily using make. This will create a shared object in the slib directory, which can be imported by Python as a module. As mentioned in the Checking section, the C++ tests will be run first. If they fail, the Python module will not be built. They do not guarantee that the Python module will work. Instead the testing stage can be used to validate the module.
Once the Python module is built, it is worth testing whether it works. This can be done with make using the command below. Unlike the C++ tests, these are Python tests that use nose to run the tests. The tests are the Python analogues of the ones used in C++ tests. They not only verify that basic command run, but that they pass with correct results only.
After building and testing, it is time to install. Using make, the command below will install the module in the identified Python’s site-package folder allowing for importing this module using that Python.
There are a few additional options regarding cleaning. It is possible to clean all build intermediates (including CMake generated files) leaving only the final build products. This is done by calling as below.
If it is desirable to eliminate the build products as well as all intermediates, then the call below can be used.