may 9-11, 2004swarmfest, cscs, university of michigan 1 jesevol pietro terna [email protected]...
Post on 20-Dec-2015
214 views
TRANSCRIPT
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
1
jES
evol
Pietro Terna [email protected]
Department of Economics and Finance “G.Prato”
University of Torino - Italy
Evolving a simulated system of enterprises with jESevol and Swarm
web.econ.unito.it/terna
web.econ.unito.it/terna/jes
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
2
_jE
S->
jES
let a
nd jE
Sev
ol _______________________________________
jES jESlet and jESevol
_______________________________________
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
3
From jES …
jVE
->jE
S
java Enterprise Simulator
… to jESlet (with a didactic goal) and …
… to jESevol, to simulate an evolving system of enterprises
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
4
_ove
rvie
w
_______________________________________
Overview
_______________________________________
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
5
over
view
1
Overview 1/2
From jES (our java Enterprise Simulator), we have derived jESevol, or “Evolutionary java Enterprise Simulator”. jES is a large Swarm-based package[1] aimed at building simulation models both of actual enterprises and of virtual ones. jESevol simulates systems of enterprises or production units in an evolutionary context, where new ones arise continuously and some of the old are dropped out.
Our environment is a social space with metaphorical distances representing trustiness and cooperation among production units (the social capital). The production is represented by a sequence of orders; each order contains a recipe, i.e. the description of the sequence of activities to be done by several units to complete a specific production.
[1] Download last versions of jES, jESlet and jESevol from
http://web.econ.unito.it/terna/jes
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
6
over
view
2
Overview 2/2
Two units can cooperate in the production process only if they are mutually visible in our social network. Units that do not receive a sufficient quantity of orders, as well as the ones that cannot send the accomplished orders to successive units, disappear.
New enterprises arise, in the attempt of filling the structural holes (Burt, 1992; Walker et al., 1997) of our social network.
A complex structure emerges from our environment, with a difficult and instable equilibrium whenever the social capital is not sufficient.
References
Burt R.S. (1992), Structural Holes – The Social Structure of Competition. Cambridge, MA, Harvard University Press.
Walker G., Kogut B., Shan W. (1997), Social Capital, Structural Holes and the Formation of an Industry Network, in Organization Science. Vol. 8, No. 2, pp.109-25.
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
7
evol
ving
sys
tem
We look at an incomplete production system continuously adapting itself to the reality coming from the global demand of the market …
… while new firms arise and old ones are dropped off
To produce goods, supply chains are created and modified, according to the changes in exiting firms
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
8
_jE
S b
asic
s
_______________________________________
jES basics
_______________________________________
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
9
WD
, D
W, W
DW
WD side or formalism: What to Do
DW side or formalism: which is Doing What
WDW formalism: When Doing What
Three formalisms
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
10
dict
iona
ry
unit = a productive structure; a unit is able to perform one of the steps required to accomplish an order
order = the object representing a good to be produced; an order contains technical information (the recipe describing the production steps)
recipe = a sequence of steps to be executed to produce a good
A dictionary
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
11
_A f
lexi
ble
sche
me
_______________________________________
A flexible scheme in jESevol
_______________________________________
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
12
DW
: a f
lexi
ble
sche
me
Units …
DW
… on a
toroidal space
(left and right borders and top and bottom ones are close together)
2
13
2
1
4
53
1
Each unit is able to
do a specific step …
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
13
WD
: rec
ipes
WD
with the recipes of the orders (what to do) expressed as sequences of numbers; orders with recipes are randomly generated with different lengths and structures
1 3 2 4
5 3
4 3 5 1 1
…
… of a recipe
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
14
mov
ing
reci
pes
DW and WD
2
13
2
1
4
53
11 3 2 4
1 3 2 4
moving around
among units
1 3 2 4
1 3 2 4 ?
lack of
visibilit
y
how to choose
Visibility is a metaphorical representation of trustiness and cooperation in a social network; when global visibility increases, we have more “social capital”
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
15
visi
bili
ty a
nd …
visibility changes
new units appear randomly (enterprise creation)
• with strategic relationships …
• … or alone
some units are dropped out
Visibility increases with the time (initial visibility is randomly chosen)
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
16
… b
ars
The left (blue) bar of each unit reports the number of waiting orders (do be done)
The down bar of each unit reports the number of consecutive clock ticks in which the unit has been idle
If > maxInactivity the unit is dropped out and all unsent products are lost
The right (red) bar of each unit reports the number of unsent products, due to the fact that a unit able to do the required step does not exist or is not visible
If > maxUnsentProducts the unit is dropped out and all unsent and waiting products are lost
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
17
_an
intr
oduc
tory
cas
e
_______________________________________
An introductory case, robust and fragile
_______________________________________
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
18
the
para
met
ers,
rob
ust i
ntro
duct
ory
case
potentialUnitTypes 5
unitGenerationInitialP 1
potentialUnitNumberPerType 2
newUnitGenerationP 0.0
interVisibilityMinLevel 0
increasingVisibilityStep 0.0
maxInactivity 10
maxUnsentProducts 10
max n. of types and max presence per type, here 5 * 2 with p=1
p of a new unit in each cycle, with a random type
in this basic case all units are visible and visibility does not change
we assume that an actual firm is dropped out from the market after three months of inactivity, so 10 ticks = 3 months of history
similarly …
Why 10? Our recipes have here maxStepNumber =5 and maxStepLength=2; potentially, in 10 ticks, each unit can receive an order, but only as a limit case; with this parameters the system can be exposed to a complete crash
Introductory robust case
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
19
intr
oduc
tory
cas
e: r
obus
t cas
e
1,000 ticks = 25 years of actual time
only 5 units kept alive
global/potential
PRODUCTION
final/potential
final/global
Introductory robust case
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
20
the
para
met
ers,
fra
gile
intr
oduc
tory
cas
e
potentialUnitTypes 10
unitGenerationInitialP 1
potentialUnitNumberPerType 1
newUnitGenerationP 0.0
interVisibilityMinLevel 0
increasingVisibilityStep 0.0
maxInactivity 10
maxUnsentProducts 10
Our recipes have here
maxStepNumber 10 and
maxStepLength 1
Introductory fragile case
max n. of types and max presence per type, here 10 * 1 with p=1
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
21
intr
oduc
tory
cas
e: f
ragi
le c
ase
150 ticks < 4 years of actual time
no units kept alive
global/potential
PRODUCTION
final/potential
final/global
Basic fragile case
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
22
_a s
tudy
cas
e
__________________________________________________
A study case, with 3 versions:
(i) basic,
(ii) increasing social capital,
(iii) with greater financial intervention of the banking system
__________________________________________________
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
23
the
para
met
ers,
bas
ic s
tudy
cas
e
potentialUnitTypes 5
unitGenerationInitialP 0.8
potentialUnitNumberPerType 2
newUnitGenerationP 0.8
interVisibilityMinLevel 1
increasingVisibilityStep 5
maxInactivity 10
maxUnsentProducts 10
max n. of types and max presence per type, here 5 * 2 with p=0.8
p of a new unit in each cycle, with a random type
in this study case, min visibility is 1, i.e. at least one common patch; visibility increases of 5 patches in each tick
(i) basic study case, starter file 5 in jESevol 0.3.00
Our recipes have here
maxStepNumber 5 and
maxStepLength 2
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
24
stud
y ca
se: b
asic
1,000 ticks = 25 years of actual time
a relevant variability in the number of units (social costs), with the trace of a cycle
global/potential
PRODUCTION
final/potential
final/global
(i) basic study case, starter file 5 in jESevol 0.3.00
a medium performance in term of potential production
some form of structure seems to emerge
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
25
the
para
met
ers,
incr
easi
ng s
ocia
l cap
ital
st
udy
case
potentialUnitTypes 5
unitGenerationInitialP 0.8
potentialUnitNumberPerType 2
newUnitGenerationP 0.8
interVisibilityMinLevel 1
increasingVisibilityStep 10
maxInactivity 10
maxUnsentProducts 10
max n. of types and max presence per type, here 5 * 2 with p=0.8
p of a new unit in each cycle, with a random type
in this study case, min visibility is 1, i.e. at least one common patch; visibility increases of 10 patches in each tick
(ii) Increasing social capital study case, starter file 5.2 in jESevol 0.3.00
Our recipes have here
maxStepNumber 5 and
maxStepLength 2
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
26
stud
y ca
se: i
ncre
asin
g so
cial
ca
pita
l
1,000 ticks = 25 years of actual time
a relevant variability in the number of units (social costs), but now with an evident cycle
global/potential
PRODUCTION
final/potential
final/global
a good (and increasing) performance in term of potential production
evident structures emerge
(ii) Increasing social capital study case, starter file 5.2 in jESevol 0.3.00
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
27
the
para
met
ers,
ban
k sy
stem
stu
dy c
ase
potentialUnitTypes 5
unitGenerationInitialP 0.8
potentialUnitNumberPerType 2
newUnitGenerationP 0.8
interVisibilityMinLevel 1
increasingVisibilityStep 5
maxInactivity 15
maxUnsentProducts 10
in this study case, min visibility is 1, i.e. at least one common patch; visibility is increases of 5 patches in each tick
(iii) Greater financial intervention of the banking system study case, starter file 5.3 in jESevol 0.3.00
Our recipes have here
maxStepNumber 5 and
maxStepLength 2
we assume that an actual firm is dropped out from the market after 15 ticks of inactivity, instead of 10
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
28
stud
y ca
se: b
ank
syst
em s
tudy
ca
se
1,000 ticks = 25 years of actual time
a less relevant variability in the number of units (reduced social cost)s, always with an evident cycle
global/potential
PRODUCTION
final/potential
final/global
a good performance in term of potential production
evident structures emerge
(iii) Greater financial intervention of the banking system study case, starter file 5.3 in jESevol 0.3.00
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
29
Sta
bili
ty; p
ersp
ecti
ves
Stability
Cases i, ii and iii are stable also running them for 4,000 ticks (one century)!
Short term enhancements
A lot of investigation is necessary on cases (i), (ii) and (iii) modelling explicitly the banking system, with the concurrent effects of the cases of bankruptcy in firms and banks
Using a Genetic Algorithm tool to choose units to be created at each tick and where to place them; the fitness will be generated by jESevol itself, from different points of view: the whole economic system, a specific unit, a cluster of units, …
May 9-11, 2004 SwarmFest, CSCS, University of Michigan
30
addr
ess
agai
n
web.econ.unito.it/terna
web.econ.unito.it/terna/jes
Let run case 5.2 or 5.3 at the question time!