Post on 14-Sep-2020
Netw
ork Analysis Lab
Budapest C
omplex S
ystems S
umm
er School
Skye B
ender-deMoll
skyebend@santafe.edu
http://student.bennington.edu/~skyebend
direction of talk:
hopefully a bit more of a sem
inar than a lecture
let me know
if I leave anything out anything important,
please interject examples or how
relevant to your own w
ork.
cover some very basic netw
ork concepts (review for m
any people)
provide references to more in-depth m
aterial for later research
demo netw
ork analysis and visualization software (but m
ostly to give examples of
capabilities, if it something you’d like to use, w
e can do a more in-depth tutorial later)
visualization of networks
brief examples of agent-based netw
ork models in R
EP
AS
T
network:
(some of this is lifted from
scott and some from
the rehannen paper, definite SN
bent)
basic idea of relational datasetone or m
ore sets with explicit relations betw
een their mem
bers
no
de
s, acto
rs, pe
op
le co
nn
ecte
d, tie
d, lin
ked
to e
ach
oth
er b
y ed
ge
s, arcs, tie
s
close relation to graph theory
important to distingusih w
hat kinds of ties are allowed in specific netw
ork
binary /vs. asym
metric
relation is either present or absent, no directionality
binary vs. valued/weighted
ties h
ave
a “w
eig
ht”, “co
st” or “d
istan
ce” a
ssocia
ted
with
the
m
symm
etric vs. asymm
etricedges are sym
metric, no directionality, A
<–>
Barcs have directionality, but can exist both w
ays, A –>
B and/or A
<– B
negative“d
istrust” , “a
void
an
ce”
multiple relations
possible to have multiple sets of relations for the sam
e actorscoding different kinds of data (friendship, trust, exchange)or the "sam
e" network at different points in tim
e
self loopsin som
e nets nodes are permitted to link to them
selves
- measurem
ent algorithms and stats m
a be ill-defined for some classes of nets
network data representation:
matrices - ¡not necessarily square!
elegant concept
closely related to graph theory (but read the other way)
useful for direct computation (m
atrix algebra)(ex. pow
ers of matrix gives distances on graph)
not very efficient for large sparse networks
adjacency matrix (square)
node X node , individual X
individual
affiliation matrix
individual X non-surveyed individuals
may be possible to extract square subset
individual X event
can use to construct square co-incidence matrix
1 1 1 1 1
1 2 3 4 5 6 7 8 9 0 1 2 3 4
E E E E E E E E E E E E E E
- - - - - - - - - - - - - -
1 EVELYN 1 1 1 1 1 1 0 1 1 0 0 0 0 0
2 LAURA 1 1 1 0 1 1 1 1 0 0 0 0 0 0
3 THERESA 0 1 1 1 1 1 1 1 1 0 0 0 0 0
4 BRENDA 1 0 1 1 1 1 1 1 0 0 0 0 0 0
5 CHARLOTTE 0 0 1 1 1 0 1 0 0 0 0 0 0 0
6 FRANCES 0 0 1 0 1 1 0 1 0 0 0 0 0 0
7 ELEANOR 0 0 0 0 1 1 1 1 0 0 0 0 0 0
8 PEARL 0 0 0 0 0 1 0 1 1 0 0 0 0 0
9 RUTH 0 0 0 0 1 0 1 1 1 0 0 0 0 0
10 VERNE 0 0 0 0 0 0 1 1 1 0 0 1 0 0
11 MYRNA 0 0 0 0 0 0 0 1 1 1 0 1 0 0
12 KATHERINE 0 0 0 0 0 0 0 1 1 1 0 1 1 1
13 SYLVIA 0 0 0 0 0 0 1 1 1 1 0 1 1 1
14 NORA 0 0 0 0 0 1 1 0 1 1 1 1 1 1
15 HELEN 0 0 0 0 0 0 1 1 0 1 1 1 1 1
16 DOROTHY 0 0 0 0 0 0 0 1 1 1 0 1 0 0
17 OLIVIA 0 0 0 0 0 0 0 0 1 0 1 0 0 0
18 FLORA 0 0 0 0 0 0 0 0 1 0 1 0 0 0
linked list
generally a list of nodes denoting their attributes
followed by a list of ties (nodeA
-> nodeB
, strength, etc)
less standardized
can store additional info
generally less efficient for some algorithm
s
more efficient storage for large sparse graphs
kinds of networks
treesno cycles
planar vs. hi-dimesional
bi-partite
where does netw
ork data come from
, and how is it collected?
sampling m
ethods
full network
requires entire population is known and can be bounded
query each individual about connections
snowball
when population is not know
nchoose starting set of individuals (random
ly, or from som
e other criteria) starting set is a very sm
all subset of the population of interestquery each individual for their connectionsrepeat for all new
individuals listed¡w
ill not locate isolates!starting point chosen m
ay influence findings
ego-centric sampling
pick random sam
ple of nodesask each to nam
e alterscheck for overlaps
observation"unobtrusive" observer records interactions or exchanges of interestrestricted dom
ain
previously recorded "incidental" data phone recordsbank bookshistorical docum
entsem
ail forwards
perscription records
random w
alk / link trace
choose individualrandom
ly choose outgoing connection to locate next individualrandom
ly choose outgoing connection from new
individualrepeat, w
atch for overlaps
Klovdahl, A
lden S. (1989) "U
rban Social N
etworks: S
ome M
ethodological Problem
s and P
ossibilities" in Kochren, M
anfred ed. T
he S
mall W
orld A
blex, Norw
ood NJ
200 residents in dense central core of Canaberra urban netw
ork
Menzel, H
erbert, and Katz, E
lihu (1955) "Social R
elations and Innovations in the Medical
Profession: T
he Epidem
iology of a New
Drug"
Pu
blic O
pin
ion
Qu
arte
rly 19:337-352 [C
lassic study tracking the diffusion of instances of a drug prescription among a
ne
two
rk of d
octo
rs]
artificially constructed studies of comm
unication, interaction, etc.
in lab setting or constrained environment w
here recording is possible
MIT
conference badges, email com
unication studies
computer sim
ulation data
networks can be generated using specific distribution or algorithm
makes possible the investigation of a class of netw
orks
- Skvoretz, John (1990) "B
iased Net T
heory: Approxim
ations, Sim
ulations and O
bservations" Socia
l Ne
two
rks 12:217-238
but it is surprisingly difficult to generate "realistic" social networks
often necessary to make assum
ptions about an underlying social process
- Jin, Em
ily., New
man, M
ark (2000) "From
Friendship to C
omm
unity: Modeling S
ocial N
etworks" S
anta Fe Institute R
EU
paperform
al analytical
random (E
rdos-Rényi, other distributions)
lattices (various geometries)
grid
ring
connection radius
“sma
ll wo
rld”
combination fraction of random
connections on regular substrate
W
atts-Strogatz circular substrate w
ith random rew
iring
other types of real world data archives:
gene linkagesflorence tax-roll datacom
piler graphsbiblom
etric citation networks
biological neuron networks
com
pu
ter n
etw
orks
company interlocks
genealogical data (family trees)
social science data archivesindustry/international trade dataS
TD
/AID
S sexual netw
orksA
IDS
needle sharingecological food w
ebsgeographic dataelectrical pow
er gridsm
ovie co-stars
some difficulties w
ith network data
bounding problems
if individuals can be linked by many-step connections, w
here does a network end?
respondent reliability problems
- F
erligoj, Anuska and H
lebec, Valentina., (1999) "E
valuation of social network
measurem
ent instruments
" So
cial N
etw
orks 21:111-130 [C
omparison of various social
support network questions and generators in S
lovinian high school]
computational storage problem
s
matrices go up as the
squ
are of the num
ber of nodes
most com
putations scale much w
orse
a few inks to data sets:
INS
NA
(International Netw
ork for Social N
etwork A
nalysis) data pagehttp://w
ww
.sfu.ca/~insna/IN
SN
A/data_inf.htm
l
Chicago Law
yers (ICP
SR
)http://w
ww
.icpsr.umich.edu:8080/A
BS
TR
AC
TS
/08218.xml?form
at=IC
PS
R
Police com
munication (IC
PS
R)
http://ww
w.icpsr.um
ich.edu:8080/AB
ST
RA
CT
S/02480.xm
l?format=
ICP
SR
Com
munity political sytem
s (ICP
SR
)http://w
ww
.icpsr.umich.edu:8080/A
BS
TR
AC
TS
/07092.xml?form
at=IC
PS
R
1992 Dom
estic Terrorism
Prepardness (IC
PS
R)
http://ww
w.icpsr.um
ich.edu:8080/AB
ST
RA
CT
S/06566.xm
l?format=
ICP
S
several example data sets are included w
ith UC
INE
T
- Marsden, P
eter V. (1990) "N
etwork D
ata and Measurem
ent" Annual R
eview
of Sociology 16:435-498 [D
iscussion of measurem
ent techniques, some general results
an
d b
iase
s]
basic network stats / concepts
degreein-degree -> num
ber of connections coming in, or "pointing tow
ards" a node
out-dgree -> num
ber of "outgoing connections"
both are the same in non-directed graphs
degree distribution of graph/network
poisson
po
we
r law
distance
(graph theoretic) min. num
ber of "steps" or links it take to traverse a path
can take into account weights of edges/arcs, sum
to a "cost function"
similarity/dissim
ilarity
important to be clear about w
hat weights of edges are m
easuring
"how close" - sim
ilarity (lager value means closer together)
"how far" - dissim
ilarity (larger value means farther apart)
diameter of graph
longest shortest path between all pairs of nodes
densitycom
plete graph / components / isolates
density of graph is number of edges as a fraction of m
ax. possible number
__
__L_
__
_
__
_L_
__
_
density = n(n-1)/2 for directed graph =
n(n-1)
clustering coeff (personal network density)
fraction of potential ties among neighbors w
hich are realizedordegree to w
hich acquaintances are acquainted with each other
__ number of edges in ngh. of v ___
clustering coeff = total num
ber of possible edges in ngh. of v
averaged over all vertices in graph
clustering coeff equal to density if graph is not very “neighbor-hoody”
value of 1 implies netw
ork disconnected but complete sub graphs
value of 0 implies that no neighbors know
each other
node-independent paths
number of unique (sam
e node is not used in more than one path) paths betw
een nodesused in algorithm
s for finding bridges and clusters in graphs
cores, cliques
clique is a fully-connected subgraph <exam
ple>
not so useful alone, usually the patter of clique overlaps is examined
often with a hierarchical clustering m
ethod
SINGLE-LINK HIERARCHICAL CLUSTERING (of TARO data)
1 1 1 1 1 1 1 1 2 2 1 1 2Level 8 9 0 3 2 4 5 6 9 5 6 4 7 1 1 0 1 3 2 7 8 2----- - - - - - - - - - - - - - - - - - - - - - - 2 . . . . XXX . . . . XXX . . XXX . . XXX . . 1 XXXXX XXXXXXX . . XXXXXXXXXXXXX XXXXXXXXXXX 0 XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
level indicates number of shared cliques, and X
Xs (crude dendogram
) indicate who w
ith
These data represent the relation of gift-giving (taro exchange) am
ong 22 households in a P
apuan village. Hage &
Harary (1983) used them
to illustrate a graph hamiltonian cycle.
Schw
imm
er points out how these ties function to define the appropriate persons to
mediate the act of asking for or receiving assistance am
ong group mem
bers.
reciprocity and transitivity
reciprocityfraction of A
-> B
ties for which there is a B
-> A
tie
transitivity (balance theory)triadsfraction of B
-> C
ties present when A
-> B
and A ->
C are present
<diagram
here>
status/control measures/centrality
locally central if it has lots of links to nodes in imm
ediate environment
globally central if it has position of strategic significance to network
degree centrality (local)“sta
r-ne
ss” of n
od
e norm
alized degree centrality is the degree divided by the maxim
um
possible degree expressed as a percentage
betweenness centrality (global)how
much a point lies betw
een other points on graphproportion of paths betw
een X and Z
that use Yevaluated over entire netw
ork - proportion of geodesics on graph that use Y
structural equivalence, role equivalence
find actors (nodes) that have similar roles in netw
ork
similar pattern of connections to other nodes, social positions
block models / im
age matrix
use cluster-analysis techniques to partition nodes into “equivalent” sets
possible examine pattern of set-relations as if it is a new
network
useful references-
Scott, John (1991) So
cial N
etw
ork A
na
lysis: A H
an
db
oo
k (2nd ed.) London, S
AG
E
Publications
sections of Scott's book are online (probably illegal) at
http://ww
w.analytictech.com
/mb119/tableof.htm
Boissevain, Jerem
y, Mitchell, C
lyde J. (1973) N
etw
ork a
nalysis: stu
die
s in h
um
an
inte
ractio
n The H
ague, Mouton
Marsden, P
eter V. (1990) "N
etwork D
ata and Measurem
ent" Annual R
eview of
Sociology
16:435-498 [Discussion of m
easurement techniques, som
e general results and biases]
analytic tech's (UC
INE
T's publisher) online netw
ork tutorial http://w
ww
.analytictech.com/netw
orks/topics.htm
Introduction to Social N
etwork M
ethods (online text and pdf) http://w
izard.ucr.edu/~rhannem
a/networks/text/textindex.htm
l
useful network softw
are
UC
INE
T - B
orgatti, Everett, and F
reeman
http://ww
w.analytictech.com
/
PA
JEK
- Vladim
ir Batagelj, A
ndrej Mrvar U
niversity of Ljubljana, Slovenia
http://vlado.fmf.uni-lj.si/pub/netw
orks/pajek/
PajekC
onverter http://student.bennington.edu/~
skyebend/pajekConvert.htm
GraphV
iz (N
eato/Dotty) - A
T&
T inform
ation visualization lab http://w
ww
.research.att.com/sw
/tools/graphviz/
trick for all of these is to figure out an effecient path to get from the data at hand to
desired results
usually more than one w
ay, just getting an easy sequence of operations and keeping track of w
here you are is sometim
es difficult.
UC
INE
T - http://w
ww
.analytictech.com/
Borgatti, E
verett, and Freem
an
P
C/w
ind
ow
sfree 30 day trial, then ~
$30US
for studentssort of the standard in the social netw
orks world
interface is a bit clunky (used to be cmd line)
matrix based
DL file form
at
UC
INE
T’s m
ain functions
matrix algebra
add, subtract, multiply, pow
ers, boolean operations on matrices
utility/data managem
ent functions sym
etrize, binarize, truncate, partitionQ
AP
regression / correlationestim
ate relationships between m
atrices by rearranging matrices
MD
S (M
ulti Dim
ensional Scaling) m
ethods (metric/non m
etric)represent as vectors, find low
dimensional em
bedding to represent them
with m
inimal distortion
useful as means of assessing "sim
ilarity" or visual layout with distances
dissimilarity m
atrixhierarchical clusteringdendogram
sblock m
odelingm
ost net stat and descriptive statscorrespondence analysisfactor analysis,cluster analysism
ultiple regressioncliques and com
ponentsread/w
rite to PA
JEK
format
library of demo data sets
kinmage
UC
INE
T operating m
etaphor
asks for input filename and param
eters
performs operations on m
atrix files
saves out new m
atrix files
reports results in log window
-----------
input p
ara
mete
rs ---> | | --->
outp
ut te
xt | P
RO
CE
DU
RE
| in
put d
ata
sets --->
| | ---> o
utp
ut d
ata
sets
-----------
important to be very careful about nam
ing,
during an analysis all of the output files will get confusing
online help files give explanation of procedures during use
P
AJE
K - http://vlado.fm
f.uni-lj.si/pub/networks/pajek/
Vladim
ir Batagelj and A
ndrej Mrvar U
niversity of Ljubljana, Slovenia
Pajek's U
ser interface
Pa
jek's D
raw
win
do
w[Davis’s Southern W
omen (B
i-partite) Data using 2-D
Fruchterm
an-Reingold A
lgorithm, D
efining Shapes and Colors of V
ertices with Input
File]
Pa
jek P
C/w
indows F
RE
EW
AR
E
network analysis and visualization
, now does m
ost stats as well
2d and 3d visualization
fast, comm
itment to sub quadratic algorithm
s
good spring-embedder visualization techniques (K
amada-K
awi, F
ruch-Riengold)
tree visualization
list based
can associate coordinates, colors, shapes with nodes and arcs.
time
ba
sed
ne
two
rks
most netw
ork descriptive stats
export BM
P, P
S, V
RM
L, SV
G,
Pajek operating m
etaphor
¡LOT
S of hierarchical m
enus!
data is maintained in list form
at files, which are loaded into m
emory
operations can performed on and using:
nets - collections of nodes and arcs/edges
partitions - segregate nodes into separate classes
clusters
permutations - reordering of nodes in net w
orks
hierarchies - trees relating nodes
vectors - numerical node attribute data
menus are arranged according to w
hat kind of data they take as input
results of one operation are used as input for the next
PajekC
onverter
- http://student.bennington.edu/~skyebend/pajekC
onvert.htm
PajekC
onverter is a basic utility (written in Java) for converting tab-delineated
text files into a format readable by the netw
ork analysis and visualization software P
ajek.
PajekC
onverter’s intention is to make it as sim
ple as possible to go from a spreadsheet
full of data to a network im
age
tries to take advantage of Pajek’s use of attributes for operations and visualization
take over the somew
hat confusing task of formatting input files
data can be pasted in or read in from text file
helps to generate nice look postscript and SV
G files by m
apping c
columns of the input text can be assigned to control the various aspects of the im
age
categorical data –> shapes or colors of nodes
numerical data –>
line weights, line w
idth, node size, node position
can handle time code data
GraphV
iz (N
eato/Dotty) - h
ttp://w
ww
.rese
arch
.att.co
m/sw
/too
ls/gra
ph
viz/ A
T&
T inform
ation visualization lab
unix/Xw
indows F
RE
EW
AR
E
good tree layout algorithms
spring embedder layout (K
K?)
scriptable (can be piped to, etc)excellent postscript output
Code for F
inite Autom
aton graph produced from dotty, im
age belowdigraph finite_state_machine {
rankdir=LR;
size="8,5"
orientation=land;
node [shape = doublecircle]; LR_0 LR_3 LR_4 LR_8;
node [shape = circle];
LR_0 -> LR_2 [ label = "SS(B)" ];
LR_0 -> LR_1 [ label = "SS(S)" ];
LR_1 -> LR_3 [ label = "S($end)" ];
LR_2 -> LR_6 [ label = "SS(b)" ];
LR_2 -> LR_5 [ label = "SS(a)" ];
LR_2 -> LR_4 [ label = "S(A)" ];
LR_5 -> LR_7 [ label = "S(b)" ];
LR_5 -> LR_5 [ label = "S(a)" ];
LR_6 -> LR_6 [ label = "S(b)" ];
LR_6 -> LR_5 [ label = "S(a)" ];
LR_7 -> LR_8 [ label = "S(b)" ];
LR_7 -> LR_5 [ label = "S(a)" ];
LR_8 -> LR_6 [ label = "S(b)" ];
LR_8 -> LR_5 [ label = "S(a)" ];
}
Finite A
utomaton graph produced from
dotty,
R S
tatistical Com
puting Environm
ent -http
://ww
w.r-p
roje
ct.org
/
SN
A R
outines for R - http://legba.hss.cmu.edu/R
.stuff/U
nix based FR
EE
WA
RE
with som
e network analysis tools
additional resources
INS
NA
links to social networks softw
arehttp://w
ww
2.heinz.cmu.edu/project/IN
SN
A/soft_inf.htm
-Sean E
verton's latest manual for using U
cinet, MS
Access, M
S E
xcel, Mage, and P
ajek for social netw
ork analysis: http://w
ww
.stanford.edu/group/esrg/siliconvalley/documents/netw
orkmem
o.doc
Visualization
no
n 'g
rap
h' m
eth
od
s
stats
blockmodel
degree distributions
adjacency matrix
de
nd
og
ram
s [from
Com
munity stru
cture
in so
cial a
nd b
iolo
gica
l netw
ork
s Girvan&
New
man 2001] F
IG. 4: (a) T
he friendship netw
ork from Z
achary's karate club study [25], as described in the text. Nodes associated w
ith the club adm
inistrator's faction are drawn as circles, w
hile those associated with the instructor's faction are draw
n as square
(b) The hierarchical tree show
ing the complete com
munity structure for the netw
ork. The initial split of the netw
ork into tw
o groups is in agreement w
ith the actual factions observed by Zachary, w
ith the exception that node 3 is m
isclassified.
fitting to (appropriate) arbitrary shape
shape reflects substrate or construction methods
circularlattice
shape reflects hierarchy (trees)
shape reflects spatial aspect of data (geographic)
reported comunication and nam
e recognition data for 22 bennington respondents arranged by residence
C
aida's skitter: The graph reflects 1,224,733 IP addresses and 2,093,194 IP links, (im
mediately adjacent
addresses in a traceroute-like path) of skitter data from 16 m
onitors probing approximately 932,000 destinations
spread across over 75,000 (70%) of globally routable netw
ork prefixes. Arranged radially according to
geographic coordinates
http://ww
w.caida.org/analysis/topology/as_core_netw
ork/about.xml
Multi D
imensional S
caling / Principal C
omponent A
nalysis
MD
S attem
pts to find a low-dim
ensional (2D or 3D
) representation of a high dim
ensional dataset with the m
inimum
of distortion
MD
S techniques
for most data sets there is no low
exact low dim
ensional embedding
idea is to find one with relatively low
stress
dim
ensions of MD
S are arbitrary
be careful about the kinds of inferences drawn
but if proprietary steps are correct, similar nodes should be grouped
optimization / spring em
bedder
rep
rese
nt n
etw
ork a
s a syste
m o
f we
igh
ts (no
de
s)
connected by springs (edges)
use some optim
ization technique to locate a low energy state for the system
Kam
ada-Kaw
ai
compute local m
inima for energy function and m
ove towards it
continue until low energy state is reached
Fruchterm
an-Riengold
uses modified force-field eqn
calculates individual displacement vectors
cooling function
GE
Maddition of local temperature, other m
odifications
SF
I colaboration network, from
C
om
mu
nity stru
cture
in so
cial a
nd
bio
log
ical n
etw
ork
s G
irvan
& N
ew
ma
n 2
00
1H
omless w
oman’s netw
ork, circular version
Hom
less wom
an’s network, spring version
hybridcentrality m
apped to icon size<
example in pajek>
y - axis to convey status
use attribute data as one dimesion, netw
ork data for another
time-based / dynam
ic
use animation to convey changes in social structure,
change in structure leading to positional change in layout
(change in daily conversation network of bennington social netw
ork study)http://student.bennington.edu/~
skyebend/pajekAnim
ator.htm
addition / deletion of ties
“slice” o
r “thre
ash
old
” disp
lay o
f tie stre
ng
ths
Kinm
age
Mo
ne
stary d
ata
, ea
ch p
lan
e sh
ow
s ne
two
rk at su
bse
qu
en
t time
“ph
ase
dia
gra
ms” o
r “retu
rn m
ap
s”
scores for the 250 alters in bennington study showing value at t0 against t1, and t1
against t2
¡most im
portant that technique is appropriate for data and problem!
refs:
Freem
an, Lin Visua
lizing
So
cial N
etw
orks
http://ww
w.library.cm
u.edu:7850/JoSS
/article.html
[includes many of these im
ages examples and som
e animations]
Tufte, E
. R. (1983). Th
e V
isua
l Disp
lay o
f Qu
an
titative
Info
rma
tion
. Cheshire, C
N:
Graphics P
ress.
CA
IDA
's index of network visualization tools
http://ww
w.caida.org/projects/internetatlas/viz/viztools.htm
l
The structure of
world trade of betw
een 28 OE
CD
countries in 1981 and 1992. The size of the nodes gives the volum
e of flows
in dollars (imports and exports) for each country . T
he size of the links stands for the volume of trade betw
een any tw
o countries. Colors give the regional respectively m
emberships in different trade organisations: E
C
countries (yellow), E
FTA
countries (green), USA
and Canada (blue), Japan (red), E
ast Asian C
ountries (pink), O
ceania (Australia , N
ew Z
ealand) (black).
more on agent based m
odeling and RE
PA
ST
ag
en
t ba
sed
is no
t a “cu
re a
ll”
usefull to gain a qualitative understanding of the implications of a theoretical m
odel
sometim
es conceptually easier to work w
ith and explain to outsiders
simple m
odels are the most m
eaningful
easy to get distracted into a model that is too com
plex to analyses usefully
important to rem
ember lim
itations, simplifications, and com
promises m
adediscrete vs. continuous
time &
simultaneity
interaction geometry
2¢ on good working strategy
best to get as clear as possible on details of theoretical model before coding
but also investigate capabilities of modeling fram
ework as guide to im
plementation
what are the starting conditions for the sim
ulation?
how w
ill time be im
plemented in this sim
?
what is the expected behavior, how
do we know
if it is working?
what data describe the system
, how w
ill it be collected and analyzed?
what space of param
eters will w
e want to explore?
are results general, or artifacts which are specific to this sim
implem
entation?
usefull repast capabilities to consider (I’m not trying to sell it!)
- http://repast.sourceforge.net
batch parameter files for extended sim
ulation runs
running in batch mode w
ithout the GU
I
flexible scheduling
quicktime m
ovie output
read in GIS
rasterfiles as landscapes
can
ge
t as fa
r un
de
r the
ho
od
as yo
u w
an
t, bu
t do
n’t h
ave
to
op
en
sou
rce
Netw
ork Agent-B
ased models in R
epast
repast architecture well suited to netw
ork models
networks can be used to construct arbitrary interaction relations
class co
nta
inin
g “sta
nd
ard
” ne
two
rk sub
strate
ge
ne
rato
rs
generally possible to try out multiple substrates w
ithout changing code
small library of netw
ork stats (but slow
, so extensive analysis should be done in other programs)
import and export netw
ork data files for later analysis
built-in network visualization tools for dynam
ic networks (needs w
ork)