NetworkX 2.7

Release date: 28 February 2022

Supports Python 3.8, 3.9, and 3.10

NetworkX is a Python package for the creation, manipulation, and study of the structure, dynamics, and functions of complex networks.

For more information, please visit our website and our gallery of examples. Please send comments and questions to the networkx-discuss mailing list.

Highlights

This release is the result of 7 months of work with over 166 pull requests by 33 contributors. Highlights include:

Warning

Hash values observed in outputs of weisfeiler_lehman_graph_hash have changed in version 2.7 due to bug fixes. See gh-4946 for details. This means that comparing hashes of the same graph computed with different versions of NetworkX (i.e. before and after version 2.7) could wrongly fail an isomorphism test (isomorphic graphs always have matching Weisfeiler-Lehman hashes). Users are advised to recalculate any stored graph hashes they may have on upgrading.

  • Dropped support for Python 3.7.

  • Added the Asadpour algorithm for solving the asymmetric traveling salesman problem: asadpour_atsp.

  • Added the Louvain community detection algorithm: louvain_communities and louvain_partitions

  • Removed all internal usage of the numpy.matrix class, and added a FutureWarning to all functions that return a numpy.matrix instance. The numpy.matrix class will be replaced with 2D numpy.ndarray instances in NetworkX 3.0.

  • Added support for the scipy.sparse array interface. This includes to_scipy_sparse_array and from_scipy_sparse_array. In NetworkX 3.0, sparse arrays will replace sparse matrices as the primary interface to scipy.sparse. New code should use to_scipy_sparse_array and from_scipy_sparse_array instead of their matrix counterparts. In addition, many functions that currently return sparse matrices now raise a FutureWarning to indicate that they will return sparse arrays instead in NetworkX 3.0.

  • Added generic dtype support to to_numpy_array. This adds support for generic attributes, such as adjacency matrices with complex weights. This also adds support for generic reduction functions in handling multigraph weights, such as mean or median. Finally, this also includes support for structured dtypes, which enables the creation of multi-attribute adjacency matrices and replaces the less generic to_numpy_recarray.

  • Added support for computing betweenness centrality on multigraphs

  • Added support for directed graphs and multigraphs to greedy_modularity_communities.

GSoC PRs

We added the work from four Google Summer of Code projects:

Improvements

  • [#4740] Add the Asadpour algorithm for solving the asymmetric traveling salesman problem.

  • [#4897] Improve the validation and performance of nx.is_matching, nx.is_maximal_matching and nx.is_perfect_matcing.

  • [#4924] Fix handling of disconnected graphs whne computing nx.common_neighbor_centrality.

  • [#4929] Add Louvain community detection.

  • [#4946] Add Weisfeiler-Lehman hashing subgraph hashing.

  • [#4950] Add an n_communities parameter to greedy_modularity_communities to terminate the search when the desired number of communities is found.

  • [#4965] and [#4996] Fix handling of relabeled nodes in greedy_modularity_communities.

  • [#4976] Add betweenness centrality for multigraphs.

  • [#4999] Fix degree_assortativity_coefficient for directed graphs.

  • [#5007] Add support for directed graphs and multigraphs to greedy_modularity_communities.

  • [#5017] Improve implementation and documentation of descendants and ancestors

  • [#5019] Improve documentation and testing for directed acyclic graph module.

  • [#5029] Improve documentation and testing of descendants_at_distance.

  • [#5032] Improve performance of complement_edges.

  • [#5045] Add geometric_edges to the nx namespace.

  • [#5051] Add support for comment characters for reading data with read_edgelist.

  • [#5052] Improve performance and add support for undirected graphs and multigraphs to transitive_closure.

  • [#5058] Improve exception handling for writing data in GraphML format.

  • [#5065] Improve support for floating point weights and resolution values in greedy_modularity_communities.

  • [#5077] Fix edge probability in fast_gnp_random_graph for directed graphs.

  • [#5086] Fix defect in lowest_common_ancestors.

  • [#5089] Add find_negative_cycle for finding negative cycles in weighted graphs.

  • [#5099] Improve documentation and testing of binary operators.

  • [#5104] Add support for self-loop edges and improve performance of vertex_cover.

  • [#5121] Improve performance of *_all binary operators.

  • [#5131] Allow edge_style to be a list of styles when drawing edges for DiGraphs.

  • [#5139] Add support for the scipy.sparse array interface.

  • [#5144] Improve readibility of node_classification functions.

  • [#5145] Adopt math.hypot which was added in Python 3.8.

  • [#5153] Fix multipartite_layout for graphs with non-numeric nodes.

  • [#5154] Allow arrowsize to be a list of arrow sizes for drawing edges.

  • [#5172] Add a nodes keyword argument to find_cliques to add support for finding maximal cliques containing only a set of nodes.

  • [#5197] Improve resistance_distance with advanced indexing.

  • [#5216] Make omega() closer to the published algorithm. The value changes slightly. The niter parameter default changes from 1->5 in lattice_reference() and from 100->5 in omega.

  • [#5217] Improve performance and readability of betweenness_centrality.

  • [#5232] Add support for None edge weights to bidirectional Djikstra algorithm.

  • [#5247] Improve performance of asynchronous label propagation algorithm for community detection, asyn_lpa_communities.

  • [#5250] Add generic dtype support to to_numpy_array.

  • [#5285] Improve karate_club_graph by updating to the weighted version from the original publication.

  • [#5287] Improve input validation for json_graph.

  • [#5288] Improve performance of strongly_connected_components.

  • [#5324] Add support for structured dtypes to to_numpy_array.

  • [#5336] Add support for the numpy.random.Generator interface for random number generation.

API Changes

  • The values in the dictionary returned by rescale_layout_dict are now numpy.ndarray objects instead of tuples. This makes the return type of rescale_layout_dict consistent with that of all of the other layout functions.

  • A FutureWarning has been added to google_matrix to indicate that the return type will change from a numpy.matrix object to a numpy.ndarray in NetworkX 3.0.

  • A FutureWarning has been added to attr_matrix to indicate that the return type will change from a numpy.matrix object to a numpy.ndarray object in NetworkX 3.0.

  • The is_*_matching functions now raise exceptions for nodes not in G in any edge.

Deprecations

  • [#5055] Deprecate the random_state alias in favor of np_random_state

  • [#5114] Deprecate the name kwarg from union as it isn’t used.

  • [#5143] Deprecate euclidean in favor of math.dist.

  • [#5166] Deprecate the hmn and lgc modules in node_classification.

  • [#5262] Deprecate to_scipy_sparse_matrix and from_scipy_sparse_matrix in favor of to_scipy_sparse_array and from_scipy_sparse_array, respectively.

  • [#5283] Deprecate make_small_graph and make_small_undirected_graph from the networkx.generators.small module.

  • [#5330] Deprecate to_numpy_recarray in favor of to_numpy_array with a structured dtype.

  • [#5341] Deprecate redundant info.

Merged PRs

A total of 166 changes have been committed.

  • Support comments=None in read/parse edgelist (#5051)

  • Add see also refs to de/stringizers in gml docstrings. (#5053)

  • Add weisfeiler lehman subgraph hashing (#4946)

  • Deprecate random_state decorator (#5055)

  • Bug fix for issue #5023 : corner-case bug in single_source_dijkstra (#5033)

  • More informative GraphML exceptions (#5058)

  • Minor updates to tutorial.rst and add docstring for data method of nodes/edges (#5039)

  • Document geometric_edges and add it to main namespace (#5045)

  • Fix small typo in trophic_levels documentation (#5087)

  • Refactor transitive_closure (#5052)

  • Fix fast_gnp_random_graph for directed graphs (issue #3389) (#5077)

  • Get number of edges by calling the proper method (#5095)

  • Update mentored projects section in docs (#5056)

  • Parametrize shortest path node-checking tests. (#5078)

  • Create FUNDING.yml

  • Deprecate union name param (#5114)

  • Update FUNDING.yml

  • vertex_cover: Added support for self-loop nodes (#5104)

  • Update core dev team (#5119)

  • Faster operators in algorithms/operators/all.py (#5121)

  • DOC: Add links to proposals for completed projects (#5122)

  • Consistent return type in dictionary output of rescale_layout and rescale_layout_dict (#5091)

  • Change exception varname e to err (#5130)

  • minor tweaks in assortativity docs and code (#5129)

  • Allow edge style to be a list of styles for DiGraphs (#5131)

  • Add examples and minor documentaion refactor for operators/binary.py (#5099)

  • Improve random graphs test suite for gnp generators (issue #5092) (#5115)

  • Add note about checking for path existence to all_simple_paths. (#5059)

  • Fix message of raised exception in decorators. (#5136)

  • Refactor linestyle test for FancyArrowPatches. (#5132)

  • Drop Py37 (#5143)

  • Use math.hypot (#5145)

  • Add pyupgrade to pre-commit (#5146)

  • Test on Python 3.10 (#4807)

  • Use black 21.9b0 (#5148)

  • Use sphinx 4.2 (#5150)

  • Update example requirements (#5151)

  • Update nx_pylab drawing edge color and width tests (#5134)

  • Refactor node_classification to improve conciseness and readability (#5144)

  • Add temporary pyparsing pin to fix CI. (#5156)

  • Add option for arrowsize to be a list (#5154)

  • List policies (#5159)

  • Bugfix for issue 5123 (#5153)

  • Test scipy and pandas on py3.10 (#5174)

  • Deprecate hmn and lgc modules from the node_classification package (#5166)

  • Rm passing ax.transOffset to LineCollection. (#5173)

  • Add a function to find the negative cycle using bellman_ford (#5089)

  • Add a Q&A to the contributor FAQ about algorithm acceptance policy. (#5177)

  • DOC: Fix typo in docs for weighted shortest paths (#5181)

  • Revert “Add temporary pyparsing pin to fix CI. (#5156)” (#5180)

  • Only compute shortest path lengths when used (#5183)

  • Add Mypy type checking infrastructure (#5127)

  • xfail pydot tests. (#5187)

  • Remove unused internal solver from algebraicconnectivity (#5190)

  • Remove check/comment for scipy 1.1 behavior. (#5191)

  • Test on Python 3.10 (#5185)

  • Add regression test for ancestors/descedants w/ undir. G. (#5188)

  • Rm internal function, use advanced indexing instead. (#5197)

  • Fix missing import + tests in laplacian fns. (#5194)

  • Investigate pre-release test failures (#5208)

  • Rm assertion method in favor of assert statements. (#5214)

  • Remove unused variable in mycielski.py (#5210)

  • used queue instead of ordinary list (#5217)

  • Add FutureWarning about matrix->array output to google_matrix (#5219)

  • A few np.matrix cleanups (#5218)

  • Rm internal laplacian in favor of laplacian_matrix. (#5196)

  • [MRG] Create plot_subgraphs.py example (#5165)

  • Add traveling salesman problem to example gallery (#4874)

  • Fixed inconsistent documentation for nbunch parameter in DiGraph.edges() (#5037)

  • Compatibility updates from testing with numpy/scipy/pytest rc’s (#5226)

  • Replace internal close fn with math.isclose. (#5224)

  • Fix Python 3.10 deprecation warning w/ int div. (#5231)

  • Touchups and suggestions for subgraph gallery example (#5225)

  • Use new package name (#5234)

  • Allowing None edges in weight function of bidirectional Dijkstra (#5232)

  • Add an FAQ about assigning issues. (#5182)

  • Update dev deps (#5243)

  • Update minor doc issues with tex notation (#5244)

  • Minor changes to speed up asynchronous label propagation for community detection. (#5247)

  • Docstrings for the small.py module (#5240)

  • Use scipy.sparse array datastructure (#5139)

  • Update sphinx (#5272)

  • Update year (#5273)

  • Update extra dependencies (#5263)

  • Update gexf website link in documentation (#5275)

  • Update numpydoc (#5274)

  • Initial setup of lazy_import functions. (#4909)

  • Deprecate scipy sparse matrix conversion functions (#5262)

  • Fix lowest_common_ancestors (issue #4942) (#5086)

  • Make small graph generator node test more specific. (#5282)

  • Use from_dict_of_lists instead of make_small_graph in generators.small (#5267)

  • Refactor to_numpy_array with advanced indexing (#5250)

  • Fix: Update louvain_partitions for threshold (update mod to new_mod in each level) (#5284)

  • Add exception for unconnected graph (#5287)

  • Fixing Tarjan’s strongly connected components algorithm implementation to have O(|E|+|V|) time complexity instead of O(|V|^3). (#5288)

  • Add weights to karate club graph (#5285)

  • Fix functions appearing in variables __all__ but not in docs for NX2.7 (#5289)

  • Update to stable version of black (#5296)

  • Add FutureWarning to attr_matrix to notify users of return type change (#5300)

  • DOC: change status to accepted for NXEP2, add resolution (#5297)

  • Update test requirements (#5304)

  • Update scipy (#5276)

  • DOC: Update documentation to include callables for weight argument (#5307)

  • Update pygraphviz (#5314)

  • Document default dtype in to_numpy_recarray docstring. (#5315)

  • Rm unused AbstractSet. (#5317)

  • Deprecate make_small_graph and make_small_undirected_graph (#5283)

  • Update draw_ docstrings with usage examples (#5264)

  • More numpy.matrix cleanups for NX2.7 (#5319)

  • MAINT: Cleanup assortativity module, remove unused variables (#5301)

  • Add informative exception for drawing multiedge labels. (#5316)

  • Potential resolution to full paths to functions in docs (#5049)

  • MAINT: Cleanup link analysis module, remove unused variables (#5306)

  • Use pytest-mpl (#4579)

  • Keep omega within [-1, 1] bounds (#5216)

  • Add support for finding maximal cliques containing a set of nodes (#5172)

  • MAINT: Remove unnecessary helper functions, use inbuilt methods for line graph generator (#5327)

  • sampling from dict_keys objects is deprecated. (#5337)

  • Add support for numpy.random.Generator (#5336)

  • Update matching functions for error validation and speed (#4897)

  • Update release requirements (#5338)

  • Add structured dtypes to to_numpy_array (#5324)

  • Deprecate to_numpy_recarray (#5330)

  • First pass at 2.7 release notes. (#5342)

  • Add pickle and yaml migration info (#5345)

  • Deprecate info (#5341)

  • Fix pandas warning (#5346)

  • Test on 3.11-dev (#5339)

  • Designate 2.7rc1 release

  • Bump release version

  • Update release process (#5348)

  • Update mentored project info with the expected time commitment (#5349)

  • Use np.random.default_rng in example + other updates. (#5356)

  • Remove stuff conda doesn’t support (#5361)

  • Fix spiral_layout when equidistant=True (#5354)

  • Fix docs (#5364)

Contributors

  • Will Badart

  • Ross Barnowski

  • Mathieu Bastian

  • Martin Becker

  • Anutosh Bhat

  • Alejandro Candioti

  • Divyansh

  • Andrew Eckart

  • Yossi Eliaz

  • Casper van Elteren

  • Simone Gasperini

  • Daniel Haden

  • Leo Klarner

  • Andrew Knyazev

  • Fabrizio Kuruc

  • Paarth Madan

  • Jarrod Millman

  • Achille Nazaret

  • NikHoh

  • Sultan Orazbayev

  • Dimitrios Papageorgiou

  • Aishwarya Ramasethu

  • Ryuki

  • Katalin Schmidt

  • Dan Schult

  • Mridul Seth

  • Cirus Thenter

  • James Trimble

  • Vadim

  • Hnatiuk Vladyslav

  • Aaron Z

  • eskountis

  • kpberry