pyGAlib is a library to generate and study graphs and complex networks in Python. It treats networks as adjacency matrices in order to take advantage of faster NumPy array manipulations. The library is easy to install, use, modify and extend.
Main features...
- networks represented as adjacency matrices (rank-2 ndarrays);
- extensive use of NumPy for high performance over pure Python code;
- Some functions further boosted using Numba;
- transparent and flexible: find which part of the code is doing what;
- Python 3 compatible.
... and some limitations:
- management of large networks due to NumPy dependence,
- no graph visualization tools.
Installation of pyGAlib is simple, only the pip package manager is needed. To check whether pip is installed, open a terminal and type:
pip --help
NOTE: If you use Anaconda (or any other third-party package manager), we recommend to install the dependencies (python>=3.6, numpy>=1.6, scipy and numba) into the target environment using Anaconda before installing pyGAlib. Otherwise,
pipwill download and install those packages directly from PyPI as well, and you won't be able to manage them through Acanconda.
pyGAlib is registered in the official Python Package Index, PyPI . To install, open a terminal window and type:
python3 -m pip install galib
To confirm the installation, open an interactive session (e.g., IPython or a Notebook) and try to import the library by typing import galib.
If you have git installed, you may like to install pyGAlib directly from its GitHub repository. Open a terminal and type:
python3 -m pip install git+https://github.com/gorkazl/pyGAlib.git@master
This will only download and install the package (files in "src/galib/") into your current environment. If desired, additional files of the repository (e.g. the examples in the Examples/ folder) should be downloaded manually. You can choose to install the version in another branch by replacing the '@master' at the end of the command by '@branchname' of the desired branch.
If you want to install pyGAlib such that you can make changes to it "on the fly" then, visit its GitHub repository https://github.com/gorkazl/pyGAlib/, select a branch and then click on the green "<> Code" button on the top right and select "Download ZIP" from the pop-up menu. Once downloaded, move the zip file to a target folder (e.g., "~/Documents/myLibraries/") and unzip the file. Open a terminal and cd to the resulting folder, e.g.,
cd ~/Documents/myLibraries/pyGAlib-master/
Once on the path (make sure it contains the pyproject.toml file), type:
python3 -m pip install -e .
Do not forget the "." at the end which means "look for the pyproject.toml file in the current directory." This will install pyGAlib such that every time changes are made to the package (located in the path chosen), these will be inmediately available. You may need to restart the IPython or Jupyter notebook session, though.
The library is organised in the following modules:
- metrics.py: Common graph metrics (degrees, clustering, graph distance, etc)
- models.py: Generation of synthetic networks and randomization.
- tools.py: Miscelaneous helper functions.
- metrics_numba.py: Uses the Numba package to accelerate calculation of some metrics.
- models_numba.py: Uses the Numba package to accelerate generation of some graph models.
- extra.py: Additional measures and functionalities related to network analysis.
Since pyGAlib depends on NumPy, it is recommended to import NumPy first. Although this is not necessary for loading pyGAlib, NumPy functionalities and array manipulation will be often needed. Try importing pyGAlib:
>>> import numpy as np
>>> import galib
NOTE: Importing galib imports also all functions in module metrics.py into its namespace. The rest of modules are imported separately. Therefore, if the import is relative those functions can be called as, e.g.,
>>> import galib
>>> ...
>>> deg = galib.Degree(net)
>>> C, Cnodes = galib.Clustering(net)
See that we did not have to call
galib.metrics.Degree(net). In the case of an absolute import (using an asterisk*) all functions in metrics.py are imported to the base namespace:
>>> from galib import *
>>> ...
>>> deg = Degree(net)
>>> C, Cnodes = Clustering(net)
Let's generate a random graph following the Erdos-Renyi model, G(N,p), with N = 100 nodes and link probability p = 0.1:
>>> import galib
>>> import galib.models
>>> N = 100; p = 0.1
>>> net = galib.models.ErdosRenyiGraph(N, p, directed=False)
Here, net is the adjacency matrix of the random graph represented as a 2-dimensional NumPy array. Let's calculate some basic properties.
>>> galib.Density(net)
0.09838383838383838
As expected, the density of an Erdos-Renyi random graph is close to the p = 0.1 value given. We now calculate the degree of every node:
>>> deg = galib.Degree(net)
>>> deg
array([10, 7, 10, 10, 11, 7, 5, 11, 13, 12, 14, 13, 8, 10, 9, 8, 7,
10, 11, 9, 11, 11, 8, 10, 5, 9, 13, 10, 13, 12, 12, 11, 11, 7,
13, 11, 7, 10, 10, 6, 12, 10, 6, 10, 7, 6, 9, 10, 9, 9, 7,
9, 8, 13, 10, 9, 7, 7, 11, 8, 13, 6, 7, 12, 14, 6, 5, 11,
5, 12, 14, 14, 13, 8, 7, 12, 4, 19, 9, 13, 7, 10, 15, 15, 4,
9, 7, 12, 7, 8, 12, 4, 11, 12, 6, 13, 6, 12, 16, 12])
The degree is returned as a numpy array of rank 1, of integer type. The function Clustering returns both the global clustering coefficient and the local clustering of every node:
>>> C, Cnodes = galib.Clustering(net)
>>> C
0.096051227321238
>>> Cnodes
array([0.13333333, 0.04761905, 0.04444444, 0.08888889, 0.10909091,
0.19047619, 0.3 , 0.12727273, 0.08974359, 0.06060606,
... ... ...
... ... ...
0.04545455, 0. , 0.10909091, 0.13636364, 0. ,
0.07692308, 0.13333333, 0.09090909, 0.08333333, 0.10606061])
We compute the pair-wise graph distance between all the nodes using the Floyd-Warshall algorithm, which is returned as a matrix dij (numpy array of rank 2):
>>> dij = galib.FloydWarshall(net)
>>> avlen = (dij.sum() - dij.trace()) / (N*(N-1))
>>> avlen
2.248080808080808
NOTE: For calculating the distance matrix of larger networks please use the version of the function located in the module metrics_numba.py. Here, the function
FloydWarshall_Numba()works the same but makes use of the Numba library to significantly speed the calculation.
Most network generators and graph metrics in pyGAlib work with directed graphs as well. Check for the optional parameter directed. Following the example above, we generate a directed Erdos-Renyi graph and calculate its input and output degrees for every node:
>>> net = galib.models.ErdosRenyiGraph(N, p, directed=True)
>>> galib.Density(net)
0.10272727272727272
>>> indeg, outdeg = galib.Degree(net, directed=True)
>>> indeg
array([17, 7, 9, 8, 11, 10, 9, 8, 13, 13, 5, 9, 13, 9, 10, 10, 13,
10, 9, 9, 7, 11, 13, 10, 4, 15, 11, 11, 10, 6, 6, 8, 8, 8,
11, 8, 4, 12, 8, 13, 13, 14, 12, 5, 6, 5, 16, 12, 5, 10, 9,
13, 8, 9, 7, 8, 13, 14, 9, 18, 7, 11, 5, 4, 12, 8, 8, 10,
7, 9, 15, 12, 14, 9, 15, 11, 13, 12, 15, 10, 11, 11, 15, 7, 10,
13, 7, 14, 9, 16, 11, 11, 6, 18, 7, 4, 14, 12, 12, 10])
>>> outdeg
array([ 9, 10, 7, 9, 12, 9, 19, 9, 11, 16, 11, 12, 11, 15, 11, 6, 9,
8, 11, 12, 9, 13, 9, 8, 11, 6, 7, 11, 11, 12, 10, 8, 11, 12,
10, 12, 13, 8, 18, 11, 8, 13, 10, 8, 10, 10, 11, 8, 11, 11, 11,
10, 11, 10, 9, 12, 6, 10, 7, 10, 10, 11, 15, 12, 11, 7, 10, 8,
5, 11, 7, 11, 13, 8, 5, 6, 13, 11, 10, 13, 7, 6, 13, 11, 8,
8, 10, 6, 10, 9, 12, 15, 11, 9, 15, 11, 7, 8, 11, 10])
Since GAlib is based on NumPy arrays, saving and reading of adjacency matrices,
as well as any other output of GAlib functions, can be performed using the
usual data I/O functionalities of NumPy. See for example the documentation for functions: loadtxt(), savetxt(), load(), save() and savez(). The tools.py module in pyGAlib provides also some data conversion functionalities.
While working in an interactive session, after importing a module, the built-in help() function will show further details:
>>> help(modulename)
The help for galib (help(galib)) shows the general summary of the package and a list of all the modules in the library. The help for each module, e.g., help(galib.metrics) or help(galib.models) will display module specific information and a list of all the functions in the module.
For further details regarding each function, type:
>>> help(galib.modulename.functionname)
IPython and Jupyter notebook users, the help command is replaced by a question mark after the module's or function's name, e.g.:
>>> modulename?
>>> functionname?
For questions, bug reports, etc, please write to gorka@Zamora-Lopez.xyz, or open an issue in GitHub.
See the TODO.md file in the GitHub Repository. Collaborations to extend pyGAlib are welcome. If you have experience using scipy.sparse, developing community detection methods or coding graph visualization, please, please, contact me.
Copyright (c) 2018, Gorka Zamora-López, <gorka@Zamora-Lopez.xyz>
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this software except in compliance with the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
Note: Please, use the logos provided in the Branding/ folder whenever possible.
- New functions added:
- Support functions
is_directed()andis_weighted()added to metrics.py module, to facilitate working with weighted graphs.
- Support functions
- Support to generate and randomize weigthed networks added. New functions included to models.py module:
SeedRandomWeights()adds random weights (sampled from a distribution of choice) to the links of an existing (di)graph.ShuffleWeights()conserves the links of a (di)graph in-place, but randomly reassigns their weights.
- "Syntactic sugar" functions were added to generate some common random / weighted graphs (models.py):
ErdosRenyiGraph_Like()generates an Erdös-Rényi (random) graph of same size and link probability as a given input network. Optionally, it also seeds random weights to the links, from a distribution of choice.RandomGraph_Like()generates a random graph of same size and number of links as a given input network. Optionally, it also seeds random weights to the links, from a distribution of choice.WeightedErdosRenyiGraph()generates a weighted Erdos-Renyi graph with link weights sampled from a distribution of choice.WeightedRandomGraph()generates a weighted random graph with link weights sampled from a distribution of choice.
- New example notebooks added:
- RichClub_Undirected.ipynb shows how to identify the presence of a rich-club in empirical (di)graphs.
- WeightedGraphs_Intro.ipynb illustrates how to use the new functions to work with weighted (di)graphs.
- Minor changes and bug fixes:
- In models.py module, function renamed from
ModularHeterogeneousGraph()toModularGraph(). - Bug fix in
ErdosRenyiGraph()function that prevented addition of self-loops even withselfloops = Trueoption set. outdtypeoption removed from graph generation functions in models.py module. For consistency, all graph generators return 2D arrays ofnp.uint8type, for the binary cases andnp.float64for the weighted (di)graphs.
- In models.py module, function renamed from
Stable version 2.0 checked, validated and released.
- Python 2 support has been dropped. Only Python 3 compatibility will be developed and maintained from now on.
- The library has been reshaped to be compliant with the modern PyPA specifications.
- Hatch was chosen as the tool to build and publish the package. See the pyproject.toml file.
- Bug fixes to adapt to the various changes in Python and NumPy since last release.
- Sample and validation scripts in the "Examples/" folder revised and adapted to recent changes in Python and NumPy.
See the file CHANGELOG.md for a complete history of changes.