1 / 32

Large Scale Social Simulation in Java and the NeISS Project

Large Scale Social Simulation in Java and the NeISS Project. Andy Turner http://www.geog.leeds.ac.uk/people/a.turner/. Overview. Introduction Modelling people movements at an individual level Demographics Traffic simulation Scaling up models Computational issues Memory handling in Java.

brendon
Télécharger la présentation

Large Scale Social Simulation in Java and the NeISS Project

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Large Scale Social Simulation in Java and the NeISS Project Andy Turner http://www.geog.leeds.ac.uk/people/a.turner/ Future Research Directions in Agent Based Modelling Workshop, Leeds, UK, 2010-06-15

  2. Overview • Introduction • Modelling people movements at an individual level • Demographics • Traffic simulation • Scaling up models • Computational issues • Memory handling in Java

  3. Introduction • GENESIS • Generative e-Social Science for Socio-Spatial Simulation • ESRC funded • A second phase research node of the UK National Centre for e-Social Science • http://www.genesis.ucl.ac.uk/ • NeISS • National e-Infrastructure for Social Simulation • JISC funded • Part of the Information Environment Program • http://www.neiss.org.uk/

  4. My Presentation Slides, and Notes from this meeting can all be found via the following URL: http://ur1.ca/07j0y

  5. Modelling People Movements at an Individual Level • Model Characteristics • Scale and resolution • Spatial • 2D/3D • Temporal • Agent and Environment Attributes • Constraints • Types of model • Prehistorical • Historical • Contemporary • Use restricted v Open access

  6. Contemporary Model Data • Lots exists, but much is not openly available • Census and Social Surveys • Government statistics • Traffic monitoring and surveys • Crowd sourced • Mobile devices • Telecommunication data • Navigation company data • Traffic/transport systems • Transport networks • Public transport scheduled services • Financial transactions

  7. Two types of Agent Based GENESIS models • Demographic model • Run with time steps of a day and run for years • Basic model is aspatial • Being spatially developed and incorporating migration • Traffic model • Run with time steps of seconds and run for days • Inherently spatial • Basic road only commuting model • Being developed with density constraints and traffic queuing

  8. Open Source Java • GENESIS Code Base • http://ur1.ca/07n5h • Key third party libraries • ojALGO • http://ojalgo.org/ • JFreeChart • http://www.jfree.org/jfreechart/ • Traveling Salesman OSM routing library • http://wiki.openstreetmap.org/wiki/Traveling_Salesman

  9. Result can be reproduced although models are stochastic in nature • Uses seeded psuedo-random number sequences • Results can be easily reproduced • Given the same input data and model configuration • A range of results can be generated for different random seeds • Iterators are used to go through collections of Objects during processing • However, the order in which objects are retrieved via the iterator does not have an effect in that the data after going through the iteration is the same each time.

  10. Basic Demographic Model • Deals with birth and death • Starts with an initial population • Comprised of males and females • Males and females may have different age specific fertility and mortality rates • At each step simulates: • Death • Birth • Pregnancy • Miscarriage

  11. Basic Demographic Model Detail • All living Person (Agents) are tested to see if they die at each step: • Tests are done by asking for the next number from a pseudo-random sequence and comparing the number with respective age and gender specific mortality rates • In the first simple model • Miscarriage rate was fixed for all ages, but was reasonably high • Gestation period was fixed at 266 days • At birth a single Agent is formed with a 50% chance of it being either male or female

  12. Example Output • _Year 299 • _TotalDeathsInYear 11442 • _TotalBirthsInYear 12049 • _TotalConceptionsInYear 14238 • _TotalMiscarriagesInYear 2138 • Output Directory

  13. Demographic considerations • Seeding the initial population with pregnancies • Without this there is a significant cohort of births 266 days after the simulation starts and then again 266 days after this • Over time miscarriage helps to even out the number of new births on each day • Sharply increasing Fertility probabilities at a specific age result in a secondary cohort effect • The first newborn population may tend to all have babies on the same day too • Pregnancy term is not fixed, but based on a probability distribution

  14. Example Platform • Intel(R) Core(TM) 2 Duo CPU P9400 @ 2.4GHz • 2.39 GHz, 1.95 GB RAM • Ran from within Netbeans 6.7.1 • Java opts • –Xmx512m –Xms512m

  15. Example Run Simulation Parameters • _MaximumNumberOfAgents = 1000000 • _MaximumNumberOfAgentsPerAgentCollection = 10000 • InitialFemalePopulation = 1000 • InitialMalePopulation = 1000 • TotalYears 300

  16. Example Run Data • No input data • All data generated by simulation model • 1162563 Files • 1173895 Directories • Size 1.92 GB (2067216392 bytes) • Size on disk 4.48 GB (4819922944 bytes)

  17. GENESIS Traffic Modelling • Agents positioned in a confined region on a Euclidean 2D plane • First Agents made to move around this randomly by repositioning at each time tick • Next the concept of destinations was developed • Rather than necessarily having a different destination at each time tick, an Agent might be assigned a destination beyond its maximum range of movement in a time tick

  18. Spatial resolutions • Fixed decimal place type precision high resolution raster network • Movement on network also uses fixed decimal place precision • Reporting done for lower resolution rasters • Both vectors of movement and population density grid outputs

  19. Model Libraries • Generic • http://ur1.ca/07ne2 • Raster • Grids • http://ur1.ca/07ndw • Vector • http://ur1.ca/07nel • Agent data

  20. Commuting model for Leeds seeded with UK Census data • Data • 2001 UK census special travel statistics data • home origin and work destination at an Output Area (OA) level • There are over 200,000 OA in the UK • Open Street Map (OSM) Road data • http://www.openstreetmap.org/ • Key 3rd party library • Traveling Salesman OSM routing library • http://wiki.openstreetmap.org/wiki/Traveling_Salesman

  21. Example output • Output Directory

  22. Computational Issues • These models are computationally demanding both in terms of the processing and the data storage requirements • To reduce compute times then parallelisation is key • MPJ Express • For large scale simulations then there are few computers with enough fast access memory • Some way of swapping data with slower access (usually persistent disk) memory is wanted • Some Java Memory Handling slides coming up...

  23. A method to get the amount of available JVM memory

  24. A method to try to prevent OutOfMemoryError being thrown

  25. A public wrapped method

  26. A simple case of handling OutOfMemoryError

  27. Agents and Data Structures • Agents • Each of these has a unique numeric ID • Have a file location • AgentCollections • These are HashSets of Agents • Have a file location • AgentIDCollections • These are HashSets of Agent numeric IDs

  28. Data Structure Settings 100000000000 1000000000 10000000 100000 1000 100 6 directory levels can store 100000000000 (one hundred thousand million) files.

  29. Considerations and Further Work • I am happy to help anyone wanting to get a copy of my code and get it running • The next big computational step is to paralellise the models • For the demographic models I am working on migration • For the traffic models I am working on implementing density and flow based constraints

  30. Acknowledgements • GENESIS is funded by the UK ESRC • RES-149-25-1078 • NeISS is funded by the UK JISC • Thanks for help and support from: • The e-Research community • The University of Leeds • School of Geography • Centre for Computational Geography

  31. Thank you http://ur1.ca/07j0y

More Related