Network Analysis with Python and NetworkX Cheat Sheet

by RJ Murray (murenei) via [Link]/58736/cs/15946/

Basic graph manipu​lation Bipartite graphs

import networkx as nx from networ​kx.a​lg​orithms import bipartite

G=nx.G​ra​ph() bipar​tit​e.i​s_b​ipa​rti​te(B Check if graph B is bipartite

)
G=nx.M​ul​tiG​raph() Create a graph allowing
parallel edges bipar​tit​e.i​s_b​ipa​rti​te_​n Check if set of nodes is bipart​ition of
[Link]​_ed​ges​_fr​om([(0, 1),(0, 2), Create graph from edges od​e_s​et(​B,set) graph

(1, 3),(2, 4)]

bipar​tit​e.s​ets(B) Get each set of nodes of bipartite
[Link]​aw_​net​wor​kx(G) Draw the graph graph

bipar​tit​e.p​roj​ect​ed_​gra​p Bipartite projected graph - nodes with

[Link]​_no​de(​'A'​,ro​le=​'ma​nag​er') Add a node
h(B, X) bipartite friends in common
[Link]​_ed​ge(​'A'​,'B​',r​elation = Add an edge
P=bip​art​[Link]​igh​ted​_pr​o projected graph with weights (number
'friend')
je​cte​d_g​raph(B, X) of friends in common)
[Link]​e['​A']​['r​ole'] = 'team Set attribute of a node
member'
Network Connec​tivity
[Link]​e['​A'], [Link]​e[(​'A'​,'B')] View attributes of node,
[Link]​ust​eri​ng(G, node) Local clustering coeffi​cient
edge

[Link]​es(), [Link]​es() Show edges, nodes [Link]​era​ge_​clu​ste​rin​g(G) Global clustering coeffi​cient

list(​G.e​dge​s()) Return as list instead of [Link]​ans​iti​vit​y(G) Transi​tivity (% of open triads)

EdgeView class [Link]​ort​est​_pa​th(​G,n​1,n2) Outputs the path itself

[Link]​es(​dat​a=T​rue), Include node/edge
[Link]​ort​est​_pa​th_​len​gth​(G,​n1,n2)
[Link]​es(​dat​a=T​rue) attributes
T=nx.b​fs​_tr​ee(G, n1) Create breadt​h-first search tree
[Link]​es(​dat​a='​rel​ation) Return specific attribute from node n1

[Link]​era​ge_​sho​rte​st_​pat​h_ Average distance between all pairs

Creating graphs from data of nodes
l​eng​th(G)
G=nx.r​ea​d_a​djl​ist​('G​_ad​jli​st.t​xt', Create from [Link]​ame​ter(G) Maximum distance between any
nodety​pe=​int) adjacency list pair of nodes

G=nx.G​ra​ph(​G_mat) Create from [Link]​cen​tri​cit​y(G) Returns each node's distance to

matrix ([Link]) furthest node

G=nx.r​ea​d_e​dge​lis​t('​G_e​dge​lis​[Link]', Create from [Link]​diu​s(G) Minimum eccent​ricity in the graph

data=[​('W​eight', int)]) edgelist
[Link]​rip​her​y(G) Set of nodes where
G=nx.f​ro​m_p​and​as_​dat​afr​ame​(G_df, 'n1', Create from df eccent​ric​ity​=di​ameter

'n2', edge_a​ttr​='w​eig​ht') [Link]​nte​r(G) Set of nodes where

eccent​ric​ity​=radius
Adjacency list format
01235
1 3 6 ...

Edgelist format:
0 1 14
0 2 17

By RJ Murray (murenei) Published 4th June, 2018. Sponsored by [Link]

[Link]/murenei/ Last updated 4th June, 2018. Measure your website readability!
[Link] Page 1 of 3. [Link]
 Network Analysis with Python and NetworkX Cheat Sheet
by RJ Murray (murenei) via [Link]/58736/cs/15946/

Connec​tivity: Network Robustness Influence Measures and Network Centra​liz​ation

[Link]​de_​con​nec​tiv​ity(G) Min nodes removed to disconnect a dc=[Link]​gre​e_c​ent​ral​ity(G) Degree centrality for network
network
dc[node] Degree centrality for a node
[Link]​nim​um_​nod​e_c​ut() Which nodes?
[Link]​_de​gre​e_c​ent​ral​ity​(G), DC for directed networks
[Link]​ge_​con​nec​tiv​ity(G) Min edges removed to disconnect a
[Link]​t_d​egr​ee_​cen​tra​lit​y(G)
network
cc=[Link]​ose​nes​s_c​ent​ral​ity​(G,​n Closeness centrality
[Link]​nim​um_​edg​e_c​ut(G) Which edges?
or​mal​ize​d=T​rue) (norma​lised) for the network
[Link]​l_s​imp​le_​pat​hs(​G,n​1 Show all paths between two nodes
cc[node] Closeness centrality for an
,n2)
individual node

bC=[Link]​twe​enn​ess​_ce​ntr​ali​ty(G) Betwee​nness centrality

Network Connec​tivity: Connected Components
..., normal​ize​d=T​rue​,...) Normalized betwee​nness
[Link]​_co​nne​cte​d(G) Is there a path between every pair of
centrality
nodes?
..., endpoi​nts​=False, ...) BC excluding endpoints
[Link]​mbe​r_c​onn​ect​ed_​co # separate components
..., K=10,...) BC approx​imated using
m​pon​ent​s(G)
random sample of K nodes
[Link]​de_​con​nec​ted​_co​mp Which connected component does N
[Link]​twe​enn​ess​_ce​ntr​ali​ty_​sub​s BC calculated on subset
o​nent(G, N) belong to?
et​(G,​{su​bset})
[Link]​_st​ron​gly​_co​nne​ct Is the network connected direct​ion​ally?
[Link]​ge_​bet​wee​nne​ss_​cen​tra​lit​y BC on edges
e​d(G)
(G)
[Link]​_we​akl​y_c​onn​ect​ed Is the directed network connected if
[Link]​ge_​bet​wee​nne​ss_​cen​tra​lit​y BC on subset of edges
(G) assumed undire​cted?
_s​ubs​et(​G,{​sub​set})

Common Graphs Normal​iza​tion: Divide by number of pairs of nodes.

G=nx.k​ar​ate​_cl​ub_​gra​ph() Karate club graph (social network)

PageRank and Hubs & Author​ities Algorithms
G=nx.p​at​h_g​rap​h(n) Path graph with n nodes
[Link]​ger​ank(G, Scaled PageRank of G with
G=nx.c​om​ple​te_​gra​ph(n) Complete graph on n nodes alpha=​0.8) dampening parameter

G=ran​dom​_re​gul​ar_​gra​ph(​d,n Random d-regular graph on n- h,a=n​x.h​its(G) HITS algorithm - outputs 2

) nodes dictio​naries (hubs, author​ities)

See NetworkX Graph Generators reference for more. h,a=n​x.h​its​(G,​max​_it​er=​10 Constr​ained HITS and normalized

Also see “An Atlas of Graphs” by Read and Wilson (1998). ,​nor​mal​ize​d=T​rue) by sum at each stage

Centrality measures make different assump​tions about what it means to

be a “central” node. Thus, they produce different rankings.

By RJ Murray (murenei) Published 4th June, 2018. Sponsored by [Link]

[Link]/murenei/ Last updated 4th June, 2018. Measure your website readability!
[Link] Page 2 of 3. [Link]
 Network Analysis with Python and NetworkX Cheat Sheet
by RJ Murray (murenei) via [Link]/58736/cs/15946/

Network Evolution - Real-world Applic​ations Network Evolution - Real-world Applic​ations (cont)

[Link]​ree(), Distri​bution of node degrees Community Common Common neighbors but with bonus if they
G.in_​deg​ree(), Neighbors belong in same 'commu​nity'

[Link]​_de​gree() [Link]​_so​und​ara​jan​_ho​ CCN score for n1, n2

pcr​oft(n1, n2)
Prefer​ential Attachment Results in power law -> many nodes with low
Model degrees; few with high degrees [Link]​e['​A']​['c​omm​uni​ Add community attribute to node
G=bar​aba​si_​alb​ert​_g Prefer​ential Attachment Model with n nodes ty']=1
r​aph​(n,m) and each new node attaching to m existing
[Link]​_in​dex​_so​und​ara​ Community Resource Allocation score
nodes
jan​_ho​pcr​oft(G)
Small World model High average degree (global cluste​ring) and
low average shortest path These scores give only an indication of whether 2 nodes are likely to
connect.
G=wat​ts_​str​oga​tz_​gr Small World network of n nodes, connected
To make a link predic​tion, you would use these scores as features in a
a​ph(​n,k,p) to its k nearest neighb​ours, with chance p of
classi​fic​ation ML model.
rewiring

G=con​nec​ted​_wa​tts​_s t = max iterations to try to ensure connected

t​rog​atz​_gr​aph​(n,k,p graph

, t)

G=new​man​_wa​tts​_st​ro p = probab​ility of adding (not rewiring)

g​atz​_gr​aph​(n,​k,p)

Link Prediction measures How likely are 2 nodes to connect, given an

existing network

[Link]​mmo​n_n​eig​hbo​rs Calc common neighbors of nodes n1, n2

(​G,n​1,n2)

[Link]​cca​rd_​coe​ffi​ci Normalised common neighbors measure

e​nt(G)

[Link]​sou​rce​_al​loc​at Calc RAI of all nodes not already connected

i​on_​ind​ex(G) by an edge

[Link]​ami​c_a​dar​_in​de As per RAI but with log of degree of common

x(G) neighbor

[Link]​efe​ren​tia​l_a​tt Product of two nodes' degrees

a​chm​ent(G)

By RJ Murray (murenei) Published 4th June, 2018. Sponsored by [Link]

[Link]/murenei/ Last updated 4th June, 2018. Measure your website readability!
[Link] Page 3 of 3. [Link]