P-SAM: A Post-Simulation Analysis Module for Agent-based Models

S. M. Niaz Arifin (1), Ryan C. Kennedy, and Gregory R. Madey

Department of Computer Science and Engineering, University of Notre Dame, IN 46556

TITLE: P-SAM: A Post-Simulation Analysis Module for Agent-Based Models

ABSTRACT: Agent-based models (ABMs) can produce large volumes of textual output, potentially in the range of hundreds of gigabytes. In most cases, these output contain inherent logical structures that can be naturally expressed in terms of abstract mathematical notions such as graphs, relations etc. It is crucial to be able to effectively analyze this vo- luminous textual output, and to produce the desired visualization with ease. Appropriate analysis and visualization also play important roles in verification & validation (V&V) of ABMs.

We have developed a software module, called P-SAM (Post-Simulation Analysis Module), to analyze and visual- ize the post-simulation output for ABMs, with special emphasis on biological simulation models. P-SAM differs from conventional statistical software tools by emphasizing the visualization part that arises from the interaction between abstract entities present in the textual output, with the goal to automate post-simulation analysis tasks for ABMs. P-SAM, though still in its current embryonic form, has been designed with the goal to handle large data files distributed over a high-performance network. To achieve this, it must be suited to take full advantages of the network's computing system, data storage system, data repositories, and visualization environments.

We describe the application of P-SAM to a biological simulation model named `LiNK' that analyzes the spread of pathogens amongst long-tailed macaque monkeys in the Indonesian island of Bali. Reported results indicate the importance of using P-SAM to perform V&V of the LiNK model. (2)

The core P-SAM architecture consists of two programs (written in Perl) called the writer and the reader. The writer takes the LiNK file as its input and serializes it (after analysis) into separate files, some of which are written in the DOT format, which allows hierarchical drawings of directed graphs. Once analysis and serialization are complete, the reader allows visualization by building the Graphical User Interface (GUI). It `reads' in the serialized DOT files, and projects the information into the GUI.

P-SAM works on relations involving relevant entities (e.g. agents) defined by the user. The visualization process enables the user to visually analyze Infection Statistics, Roaming Infection Statistics, Birth and Death Statistics, Pathogen Transmission Graphs, and Summary Statistics.

P-SAM still faces the challenges of efficiently processing hundreds of gigabytes of data and minimizing the response time for large data sets run over the campus network. In collaboration with the Center for Research Computing (CRC) at the University of Notre Dame, we currently focus on the following tasks for performance improvements:

• Preprocessing Decomposing the large LiNK file into smaller files according to similar events, then to sort and

analyze

• Profiling Using profilers (Devel::Profile and Devel::NYTProf) to measure the frequency and duration of function

calls and to collect performance data

• Code Optimization Applying certain optimization techniques after the hotspots have been identified by profiling

Once the above are implemented within the cyberinfrastructure setting offered by CRC, we expect P-SAM to be able to perform more complex, multi-run analysis involving large data sets with improved runtime, and to efficiently handle the storage, management, integration, and visualization of data produced by the LiNK model. All of these, linked by the network, would create opportunities of scholarly innovation and discoveries by the end-users. This, in turn, would allow the biologists to further validate and refine their model. We also envision P-SAM to be useful to other types of ABMs.

(1) sarifin@nd.edu

(2) See http://www.nd.edu/~macaque/ for more information

Back to Swarmfest2010_program