GraKeL is a library compatible with the project of scikit-learn
The GraKeL library requires:
- Python (>=2.7, >=3.5)
- NumPy (>=1.8.2)
- SciPy (>=0.13.3)
- Cython (>=0.27.3)
- cvxopt (>=1.2.0) [optional: lovasz]
- future (>=0.16.0) (for python 2.7)
You can install the dependencies as follows:
$ pip install extension>=extension_versionor
$ pip install -r requirements.txtTo install the development-version using pip execute:
$ pip install grakel-devTo learn how to use the GraKeL api as a user, please read the documentation on sections Introduction and A longer introduction (in case your are full of curiosity).
from grakel import GraphKernel
wl_subtree = GraphKernel(kernel=['WL', 'ST-WL'], normalize=True)-
Custom Input
H2O = [[(0, 1), (0, 2), (2, 0), (1, 0)], # Directed Graph {0: 'O', 1: 'H', 2: 'H'}] # Node Labels H3O = [[(0, 1), (0, 2), (0, 3), (3, 0), (2, 0), (1, 0)], # Directed Graph {0: 'O', 1: 'H', 2: 'H', 3:'H'}]] # Node Labels X = [H2O, H3O] # List of Graph-Like Objects
-
Download a Dataset
from grakel.datasets import fetch_dataset MUTAG = fetch_dataset("MUTAG") X = MUTAG.data # MUTAG.target contains class labels
K = wl_subtree.fit_transform(X) # len(X) x len(X): symmetricYou first need to build the package's c++ extensions locally by executing:
$ python setup.py build_ext -iTo test the package, execute:
$ nosetestsFor executing unit_tests or use a testing-interface for testing the kernel module:
$ python grakel/tests/test_kernels.py --help
usage: test_kernels.py [-h] [--verbose] [--problematic] [--slow]
[--ignore_warnings] [--dataset DATASET] [--normalize]
[--develop | --all | --main]
A test file for all kernels
optional arguments:
-h, --help show this help message and exit
--verbose print kernels with their outputs on stdout
--problematic allow execution of problematic test cases in development
--slow allow execution of slow test cases in development
--ignore_warnings ignore warnings produced by kernel executions
--dataset DATASET chose the datset you want the tests to be executed
--normalize normalize the kernel output
--develop execute only tests connected with current development
--all execute all tests
--main execute the main tests [default]
For testing graph_kernels:
$ python grakel/tests/test_graph_kernel.py --help
usage: test_graph_kernels.py [-h] [--verbose] [--problematic] [--slow]
[--normalize] [--ignore_warnings]
[--dataset DATASET] [--develop | --all | --main]
A test file for all kernels
optional arguments:
-h, --help show this help message and exit
--verbose print kernels with their outputs on stdout
--problematic allow execution of problematic test cases in development
--slow allow execution of slow test cases in development
--normalize normalize the kernel output
--ignore_warnings ignore warnings produced by kernel executions
--dataset DATASET chose the datset you want the tests to be executed
--develop execute only tests connected with current development
--all execute all tests
--main execute the main tests [default]
And for testing the Graph class:
$ python grakel/tests/test_graph.py --help
usage: test_graph.py [-h] [--verbose] [--ignore_warnings]
A test file for all `Graph` type objects
optional arguments:
-h, --help show this help message and exit
--verbose verbose outputs on stdout
--ignore_warnings ignore warnings produced by kernel executionsYou can also execute the kernel test locally through a test-main-function as
$ python -m grakel.testsTo learn how to integrate your own kernel and how to contribute to the GraKeL project, please read the sections Write your own kernel and Contributing in the Documentation, respectively.
GraKeL comes with a BSD 3-clause license (as with scikit-learn). It contains the C++ source code of BLISS (a library for graph isomorphism) which is LGPL licensed. Futhermore its optional dependency in the package of cvxopt (a tool for solving convex-optimization problems) comes with a GPL license.
If you use GraKeL in a scientific publication, please cite our paper:
@article{siglidis2018grakel,
title={GraKeL: A Graph Kernel Library in Python},
author={Siglidis, Giannis and Nikolentzos, Giannis and Limnios, Stratis and Giatsidis, Christos and Skianis, Konstantinos and Vazirgiannis, Michalis},
journal={arXiv preprint arXiv:1806.02193},
year={2018}
}