Agent-based modeling is a technique used to explore both complexity and emergence by simulating individual actors and their actions within a system. Think of systems such as the traffic in a city, or like those in financial markets where one actor can have an effect on the decisions of others until the system’s direction changes its course. In this talk, you will learn about ABMs in Python.

1. Agent-Based Modeling
& Python
By Jackie Kazil
DC, Oct. 8, 2016
1

2. 2
PSF
PyLadies
WDSDC
Data Community DC
PIFF
Byte Back

3. 3
PyData 2016 DC host and sponsor

4. 4
Class of 2019?

5. 5
Overview
1. What Agent-based modeling (& Complexity)
2. Background in Agent-based modeling tools
3. Modeling in Python using Mesa
4. Future of Mesa

6. 6

7. 7

8. 8
AGENT-BASED MODELS!
An Overview of Applications

9. 9
AGENT-BASED MODELS!
ABMs
MAS
Game theory
Cellular automata
Microsimulation
Individual-based models
Complex Simulations

10. 10
What is agent-based modeling?
Computer simulation
Consisting of agents
... interacting with one another
... in order to study an overall system

11. 11
Components of a model
Space
Agents
Time
Visualization (not required)

12. Conway’s Game of Life
Source:
Wikipedia user LucasVB,
https://commons.wikimedia.or
g/wiki/File:Gospers_glider_gun
.gif
12

13. Schelling Segregation Model
Source:
Case, “Parable of the Polygons”
http://ncase.me/polygons/
Source:
Schelling, 1971. “Dynamic
Models of Segregation.”
Journal of Mathematical
Sociology.
13

14. 14
Why ABMs?
Flows: evacuation, traffic, and customer flow
management.
Markets: stock market, shopbots and software agents,
and strategic simulation.
Organizations: operational risk and organizational design.
Diffusion: diffusion of innovation and adoption dynamics.

15. 15
Why ABMs?
ABMs capture the path as well as the solution, so one can
analyze the system’s dynamic history.
Most social processes involve spatial or network attributes,
which ABMs can incorporate explicitly.
When a model (A) produces a result (R), one has established
a sufficiency theorem, meaning R if A.

16. 16
Why ABMs?
Individual behavior is nonlinear; characterized by
thresholds, if-then rules, or nonlinear coupling
Individual behavior exhibits memory, path-dependence, or
temporal correlations, including learning and adaptation
Agent interactions are heterogeneous; can generate
network effects --> lead to deviations from predicted
aggregate behavior

17. Predator-Prey Dynamics
Source:
Wilensky, U.
(1997). NetLogo
Wolf Sheep
Predation model.
http://ccl.northwe
stern.edu/netlogo
/models/WolfShee
pPredation.
Center for
Connected
Learning and
Computer-Based
Modeling,
Northwestern
University,
Evanston, IL.
17

18. Political Dynamics
Source:
Cederman et al.
“GeoSim
Framework.”
http://www.icr.eth
z.ch/research/geo
sim
18

19. Migration Modeling
Source:
Gulden et al.
2011, “Modeling
Cities and
Displacement
through an
Agent-Based
Spatial Interaction
Model,”
Computational
Social Science
Society of America
Conference
19

20. Epidemiological Simulation
Source:
Los Alamos
National
Laboratory,
OPPIE
epidemiologica
l model.
http://www.lan
l.gov/projects/
mathematical-
computational-
epidemiology/
agent-based-m
odeling.php
20

22. 22
Open Source
ABM Library Ecosystem

23. MIMOSE (Micro-und Multilevel
Modelling Software) [53]
GAMA [28]
ABLE (Agent Building and
Learning Environment) [2]
Jason (Jason:Interpreter for
extension of AgentSpeak) [40]
MAS-SOC (Multi-Agent Simulations
for the SOCial Sciences) [51]
Brahms [11]
ECHO [22]
FLAME [25]
FLAME GPU [26]
Construct [13]
MOOSE(Multimodeling
Object-Oriented Simulation
Environment) [55]
Jade’s sim++
SimBioSys [68]
SimPack [69]
DeX [20]
LSD (Laboratory for
Simulation Development)[45]
iGen [31]
FLUXY [27]
MASS (Multi-Agent Simulation Suit)
[50]
Framsticks
SimPlusPlus [66]
MASS (Multi-Agent Simulation Suit) [50]
Framsticks
SimPlusPlus [66]
Descartes[19]
A3 / AAA (Agent Anytime Anywhere)
[1]
ADK (TryllianAgent Development
Kit)[4]
AOR Simulation[6]
Ascape [8]
BDI4Jade [9]
Cougaar[15]
CybelePro[17]
FAMOJA(Framework for Agent-based
MOdelling with JAva) [24]
ICARO-T [32]
JABM [34]
JADE
JAS-mine [38]
JASA (Java Auction Simulator API)
[39]
jES (Java Enterprise Simulator) [42]
jEcho [43]
MASON [49]
Moduleco [54]
netLogo
JIAC [64]
Sugarscape [71]
VSEit [74]
NetLogo
Spatial Modeling Environment(SME) [70]
JAMEL (Java Agent-based MacroEconomic Laboratory)
[35]
PS-I (Political Science-Identity)[60]
Altreva Adaptive Modeler [3]
oRIS [59]
SimAgent (alsosim agent) [67]
DigiHive
Mesa [52]
DALI[18]
Breve
SeSAm (Shell for Simulated Agent Systems) (fully
integrated graphical simulation environment) [63]
Cormas(Common-pool Resources and Multi-Agent
Systems) [14]
SDML (Strictly Declarative Modeling
Language) [61]
Soar
StarLogo
OpenStarLogo
StarLogo TNG
StarLogoT
SEAS (System Effectiveness Analysis
Simulation) [62]
GPU Agents [29]
GROWlab [30]
AgentSheets
VisualBots [73]
ZEUS [76]

24. MIMOSE (Micro-und Multilevel
Modelling Software) [53]
GAMA [28]
ABLE (Agent Building and
Learning Environment) [2]
Jason (Jason:Interpreter for
extension of AgentSpeak) [40]
MAS-SOC (Multi-Agent Simulations
for the SOCial Sciences) [51]
Brahms [11]
ECHO [22]
FLAME [25]
FLAME GPU [26]
Construct [13]
MOOSE(Multimodeling
Object-Oriented Simulation
Environment) [55]
Jade’s sim++
SimBioSys [68]
SimPack [69]
DeX [20]
LSD (Laboratory for
Simulation Development)[45]
iGen [31]
FLUXY [27]
MASS (Multi-Agent Simulation Suit)
[50]
Framsticks
SimPlusPlus [66]
MASS (Multi-Agent Simulation Suit) [50]
Framsticks
SimPlusPlus [66]
Descartes[19]
A3 / AAA (Agent Anytime Anywhere)
[1]
ADK (TryllianAgent Development
Kit)[4]
AOR Simulation[6]
Ascape [8]
BDI4Jade [9]
Cougaar[15]
CybelePro[17]
FAMOJA(Framework for Agent-based
MOdelling with JAva) [24]
ICARO-T [32]
JABM [34]
JADE
JAS-mine [38]
JASA (Java Auction Simulator API)
[39]
jES (Java Enterprise Simulator) [42]
jEcho [43]
MASON [49]
Moduleco [54]
netLogo
JIAC [64]
Sugarscape [71]
VSEit [74]
NetLogo
Spatial Modeling Environment(SME) [70]
JAMEL (Java Agent-based MacroEconomic Laboratory)
[35]
PS-I (Political Science-Identity)[60]
Altreva Adaptive Modeler [3]
oRIS [59]
SimAgent (alsosim agent) [67]
DigiHive
Mesa [52]
DALI[18]
Breve
SeSAm (Shell for Simulated Agent Systems) (fully
integrated graphical simulation environment) [63]
Cormas(Common-pool Resources and Multi-Agent
Systems) [14]
SDML (Strictly Declarative Modeling
Language) [61]
Soar
StarLogo
OpenStarLogo
StarLogo TNG
StarLogoT
SEAS (System Effectiveness Analysis
Simulation) [62]
GPU Agents [29]
GROWlab [30]
AgentSheets
VisualBots [73]
ZEUS [76]
http://bit.ly/wiki-abms

25. Free w/ strings attached 27
GPL 16
LGPL 14
Custom licenses 10
Generic ‘Open Source’ 7
Academic free license 4
Apache 2.0 4
‘Unknown’ 4
BSD 3
ABM licenses

26. 26
Most models are from...
netLogo (Logo) - bit.ly/abm-netlogo (~60%)
MASON (Java) - bit.ly/abm-mason
RePast (Java) -bit.ly/abm-repast

27. 27
Name 2.7 3+ Active
Dates
PyPI Description
Simx Y N 11/12-
12/14
Y Framework for discrete simulations,
optimized for parallel computing; no
built in visualization
PyCX Y Y 06/11-
03/16
N Repository of ABM examples & GUI
script for desktop visualization;
Focused on ease of writing; Mostly pure
python; not a framework
PyABM Y N 09/12-
03/14
Y Partial ABM framework, not working
Indra N Y 12/14-
Today
N ABM framework to write models similar
to Netlogo; lacks documentation;
visualization ability in the future?
Mesa N Y 09/14-
Today
Y ABM framework to build models with
repeatable components; uses browser
for visualization

28. Modeling in browser
Parable of the Polygons
(Javascript & HTML) -
ncase.me/polygons
Agent Base (Javascript) -
bit.ly/abm-ants
Agent Script (CoffeeScript) -
agentscript.org
Mesa (Python) - bit.ly/abm-mesa
28

29. 29
Mesa, ABMs in Python

30. 30
Demos of Models
Docs:
bit.ly/abm-mesa
Code:
github.com/projectmesa/mesa

31. 31
Demos of Schelling
README Standard
$ pip install -r requirements.txt
model.py - agents & model
Using model.py in jupyter
Creating browser front-end
Launching browser front-end:
$ python run.py
Open browser: 127.0.0.1:8888

32. Complex
model
32
Simple
model

33. Mesa’s Scheduling Feature
Notebook example & source: Prisoner’s dilemma
● Sequential activation, where agents are
activated in the order they were added to the
model
● Random activation, where they are activated in
random order every step
● Simultaneous activation, simulating them all
being activated simultaneously.
33

34. Mesa’s Scheduling Feature
Notebook example & source: Prisoner’s dilemma
34
Sequential
Random
Simultaneous

35. Mesa
35
Models built using Mesa

37. Mesa
37
Models built using Mesa

38. Mesa
38
Networks

39. Geo-Mesa
Mesa
39
Networks

40. Mesa
40
Econ-Mesa

41. Mesa
41
Bio-Mesa

42. Mesa
42
Other school
of thought

43. 43
Mesa Growth / Usage
Started in 2014 with 2 people. Now 26!
21 were from Sprints!
Most were not ABM experts.
Students have used it. Gov has shown interest.
ToDo list: Networks (in progress), GIS,
Parallelization of models, Front-end, Increase
usability, fix all things

44. 44
Getting Started &
Contributing
Docs & tutorial: bit.ly/abm-mesa
Look at Examples in repo
Help with ticket #223
Translate a model from NetLogo

45. 45
Things to read
Cioffi-Revilla, C. (2013). Introduction to computational social science.
Epstein, J. M., & Axtell, R. (1996). Growing artificial societies: social
science from the bottom up.
Gilbert, N., & Troitzsch, K. (2005). Simulation for the social scientist.
McGraw-Hill Education (UK).
Kupers, R., & Colander, D. (2014). Complexity & the Art of Public Policy.
Miller, J. H., & Page, S. E. (2009). Complex adaptive systems.
Simon, H. A. (1996). The sciences of the artificial.

46. 46
The End.
Questions?
@jackiekazil
github.com/projectmesa/mesa