Zed Langston
April 14, 2015
Assignment 2: Cellular Automata (CA) Modeling
Description:
This is a segregation model that shows the behavior-based settlement patterns of two different agents living within the same environment. The agents’ behavior is based on the neighborhood cell influence using “%-similar-wanted” rule parameter that sets the acceptability threshold for similar autonomous agents to self-organize into cell neighborhoods with similar agents. The default nearest neighbor (resampling) method in NetLogo are based on the Moore neighborhood (8 adjacent cells to the center cell) and the cell (patch) state over time. The result is that some agents will be unhappy and unable to occupy some cells in a neighborhood because the empty cell does not meet the individual or environmental requirements as influenced by the majority of surrounding cells, occupied by different opposite agents. The initial model (at setup) randomly assigns the spatial location to the agents (set at 1,000) that are divided into two classes (red and white). Each class has 500 similar autonomous agents that make decision about cell settlement with the goal being in a neighborhood with at least 25% of the cell neighbors being similar. The baseline “%-similar-wanted” parameter was set to 25%. The result is a 68.7 “%-similar”self-organized pattern. My assumptions for this model scenario is that “%-similar” results over 60% is not a random pattern but an emergent pattern. Movie files were originally included but were later removed due to the short run time. Below are images of the baseline (left) and finished results (right).
Questions:
1) The minimum level where self-preference leads to a segregated emergent pattern is 1%. Although the self-organization at this level of self-preference results in a fairly minor level of segregation, there is some evidence of segregation because there are some distinct groups after the local interactions among the agents and cells. The model interface baseline monitor starts at ~48% and the model stops quickly, after 6 ticks (time steps) at 61.7% neighborhood cell similarity. My assumption for this model scenario is that any result over 60% similar is a pattern where the agents self-organized at a level to produce a higher-level pattern. Below are images of the model baseline (left) and results (right).
2) The specific threshold in this model where the system operation changes from metastable to a continuous dynamic operation, for 1,000 agents, is any value above 80%. This model run used 81% for the “%-similar” parameter set point. As the number of agents is increased the threshold becomes less. At this “%-similar-wanted” parameter preference the model will stop running after ~2,000 (the automatic scale shows 2,190 ticks). At this threshold there is 57.2% similar agent occupied cells with 743 agents that are not happy because they could not find a suitable cell to occupy because of the difference (influence of) in neighbor cells. Almost 75% of the agents could not find suitable cells where their requirements were met. Below is a movie file that displays the results and images of the baseline (left) and finished results (right) that display the continuous dynamic fluctuation.
3) Changing the “%-similar-wanted” rule parameter changes the segregation patterns because it determines the resultant positioning of the agents’ settlement. As the parameter is increased the neighborhoods have increasing influence on agent settlement patterns and behavior. Many agents will settle fast because of the establishment of similarity in the overall neighborhoods. Other agents will have to move more times because of the neighborhood cell influence on the surrounding neighbor cells to find a suitable neighborhood based on the similarity (self-preference) rule. Below are the image results (emergent patterns) using 1,000 agents and 20% self-preference increments starting with 20% (top left – image 1), 40% (top right – image 2), 60% (bottom left – image 3) and 80% (bottom right – image 4). In image 1 (top left) there is some (very minor) segregation where the overall neighborhood conglomerates are smaller and more distributed. Image 2 (top right) shows the emergent pattern more clearly (it is easier to see the segregation). In images 1 and 2, it looks like the segregation (preference) pattern is almost linear in design (shape). Image 3 (bottom-left) shows larger aggregations (macro-behavior) of agents in larger similar neighborhoods. Image 4 (bottom right) gives a very clear picture of segregation where there are very distinct mostly homogenous neighborhoods especially at the bottom left and the center (green agents) and lower right, bottom-center and mid left regions (white agents) of the image. This was challenging despite several attempts because of my unfamiliarity with the NetLogo program and code required to change the self-preference parameter while “grabbing” the video. At a “%-similar-wanted” (self-preference) parameter setting of 20 the model runs for five ticks and is too short to adjust the parameter before the command finishes.
4) Increasing the “%-similar-wanted” parameter changes the nature of emergent patterns because agents (species) have to make more moves in a system with more constraints to find suitable neighborhoods (habitats). The changes in the patterns appear more organic in shape in contrast to the linear nature at a lower parameter setting. The spatial distribution also appears more condensed (more agents in a single cellular neighborhood and less neighborhoods overall with larger conglomerates at a higher parameter setting. However, after 81% the model can not find suitable neighborhood cells to meet the similar self-preference threshold as described above. Using initial settings of 1,500 agents and “%-similar-wanted” value of 70 the 9 images below cover the first 9 time steps. Image 1 (top left) has an unhappy ratio of 69.7% with 455 similar neighbors. Image 2 (top center) has an unhappy ratio of 63.7% with 545 similar neighbors. Image 3 (top right) has an unhappy ratio of 59.9% with 601 similar neighbors. Image 4 (middle left) has an unhappy ratio of 55.2% with 672 similar neighbors. Image 5 (middle center) has an unhappy ratio of 52.1% with 719 similar neighbors. Image 6 (middle right) has an unhappy ratio of 48% with 780 similar neighbors. Image 7 (Bottom left) has an unhappy ratio of 47.6% with 786 similar neighbors. Image 8 (bottom center) has an unhappy ratio of 47.2% with 792 similar neighbors. Image 9 (bottom right) has an unhappy ratio of 43.7% with 844 similar neighbors. There appears to be fewer overall specific neighborhoods that each contain more members. What this indicates is that there is a positive correlation (influence) between the number of similar neighbors and the self-preference parameter. The higher the self-preference, the more similar the neighbors are until the threshold is reached. This means at each successive time step it becomes more difficult for remaining agents to find suitable requirements (competition). Below the threshold, the unhappy ratio decreases at each time step because neighbor similarity is increasing. Below the NetLogo images are images from Excel.
5) The general conclusion is that with any amount of self-preference (a universal trait) the emergent pattern includes some level of segregation. In every environment (natural or built) there is competition for resources, habitats or micro-climates. There are going to be clusters of particular species, agents or neighborhoods because every agent has some self-preference in spatial settlement based on biological, environmental or social needs. Most organisms (agents) require more than one type of habitat, often many. In a finite environment with a 1,000 individuals only 80% can settle next to similar habitat types. With species that have a high preference (>80%), mobility must increase as most of the individuals will have to move continuously to try to find the suitable or required habitat conditions. In addition, the higher the “%-similar-wanted” parameter setting, the longer it takes (in time) for agents to find suitable neighborhoods. However, at a higher “%-similar-wanted” level, species will self-organize into larger cohesive units (neighborhoods). Below is a video that displays the results of 2,000 agents using a “%-similar-wanted” parameter set point of 70. Initially, there is 49% neighbor similarity with 84.2% unhappy neighbors. The result is an emergent pattern with a 99.4% neighbor similarity in 66 time steps.
Leave a Reply