Application independent sample free population synthesis framework
This is a heuristic method for synthesising agent populations using survey data. This was partially funded by Life Time Affordable and Tennable City Housing (LATCH) project .
This is a software for synthsising agent populations using data sources like surveys and census. Significances of this project are:
- Do not rely on sample data (microdata)
- Uses a novel framework capture population heuristics in a generic manner allowing the software to be used to generate different populations simply by changing the inputs. A feature that other sample free population synthesis methods do not have, to the best of our knowledge.
The program has two main parts:
- A framework to capture population heuristics in a generic manner
- A population synthesis algorithm that uses Iterative Proportional Fitting, Monte Carlo Sampling and Hill Climbing Optimisation.
The purpose of the framework is decoupling the algorithm and application specific heuristics by providing a structured interface. Sample free population synthesis algorithms that we have seen in the literature have heuristics hardcoded into the algorithm. This makes them hard to use in different applications other than what they are initial designed for. We avoid this problem by decoupling the heuristics from the algorithm as mentioned earlier. The algorithm implemented here acts on the constructs of the framework when synthesising the population rather than directly on the heuristics. This allows us to change the input heuristics independently for different applications without having to change the algorithm code.
Full description of the underlying framework and the algorithm is available at
doc/full algorithm.pdf. This paper will soon appear in Journal of Artificial Societies and Social Simulation (JASSS).
Data distribution - This is a categorical data distribution giving the number of persons or households (family) under different person or household categories. Usually a data distribution comes in tabular form from a survey or a census database.
Characteristic - is a distribution of households or persons in the population. e.g. number of persons by gender distribution, number of households by household sizes
Category- 'Male' and 'Female' are examples for categories in number of persons by gender population charactristic
Agent type / group type - Is a cohort of agents (persons) or groups (households) based on a combination of categories.
Agent - An agent instance belonging to some agent type
Link Type - A relationship type agents can form with other agents (e.g. marital and relative). Link type also defines maximum number of relationships that can be formed and the minimum number that must be formed by an agent. For example, given a person instance is in married category that person must from minimum of one and a maximum of one martial relationship. This relationship type can be represented as
(marrital, 1,1). Link types are defined for an agent type.
Link - An instance of a relationship between two agent instances
Link Rule - Defines a relationship that a given agent type can from with other agents types. For example,
(married, male, 25-39) person type can form marital relationships with
(married, female, 25-39) and (married, female, 15-24) person types. The link rule for this is
Group - An instance of a group
Group Rule - A mapping function to map a group of agents (a group instance) to a group type based on the agent types and their links.