152
X-Machines for Agent-Based Modeling: FLAME Perspectives
6.4 A Virtual Mall Model: Labor and Goods Market
Combined
The virtual mall model is a simple representation of interactions between
the labor and goods markets. The model uses learning to allow agents to learn
their most profitable strategies in a changing environment. Learning allows
new strategies to be produced in the market, which are not pre-coded at the
start of the experiment. The model involves four agents described below.
Malls. The Mall agents functions on a monthly cycle, at every 20 workable
days. At the start of every month, the malls use a strategy posted to
them by the environment agent and use it for 5 days. At the end of 5
days they assess its performance. The current performance is compared
to the gradient at the end of the previous month, at which point the
mall decides to adapt this new strategy for the rest of the month or
switch to the old one.
Persons. The Person agents possess an extra variable in their memory called
the learning window. If the person strategy proves to be better, the
learning window increases in length, causing the assessment period for
the person to appear at a later stage. Thus the lengths of the windows
can give us insights on whether the people were performing well or not
as well as their gradients.
Environment. This agent is responsible communicating information to the
agents. The strategies are also held in the environment and are posted
at the time the agents need to try out new strategies. This is in contrast
for keeping the strategies in the agent memories. The environment also
holds the messages for the agents which are to be used within the strategy. By keeping them here, we are able to remove the communication
dependency within the strategy gene of every agent allowing free access
for functions to be moved around.
Message counter. This agent is responsible for keeping track of the messages being sent between agents. This would allow us to see the reasons
for some of the emerging behavior of agents.
The experiment was used to test various hypotheses and understand how
learning and behavior are seen in a labor and goods market model. The simulation was started with neither the malls nor the people having knowledge
about their previous actions and the results were as follows.
Best performing malls. The malls were told to go bankrupt if their capitals
went below a bankruptcy level. Therefore, whenever this happened, the
malls were removed from the simulation. The list of functions which the
malls could include in their strategies were
X-Machines for Agent-Based Modeling: FLAME Perspectives
6.4 A Virtual Mall Model: Labor and Goods Market
Combined
The virtual mall model is a simple representation of interactions between
the labor and goods markets. The model uses learning to allow agents to learn
their most profitable strategies in a changing environment. Learning allows
new strategies to be produced in the market, which are not pre-coded at the
start of the experiment. The model involves four agents described below.
Malls. The Mall agents functions on a monthly cycle, at every 20 workable
days. At the start of every month, the malls use a strategy posted to
them by the environment agent and use it for 5 days. At the end of 5
days they assess its performance. The current performance is compared
to the gradient at the end of the previous month, at which point the
mall decides to adapt this new strategy for the rest of the month or
switch to the old one.
Persons. The Person agents possess an extra variable in their memory called
the learning window. If the person strategy proves to be better, the
learning window increases in length, causing the assessment period for
the person to appear at a later stage. Thus the lengths of the windows
can give us insights on whether the people were performing well or not
as well as their gradients.
Environment. This agent is responsible communicating information to the
agents. The strategies are also held in the environment and are posted
at the time the agents need to try out new strategies. This is in contrast
for keeping the strategies in the agent memories. The environment also
holds the messages for the agents which are to be used within the strategy. By keeping them here, we are able to remove the communication
dependency within the strategy gene of every agent allowing free access
for functions to be moved around.
Message counter. This agent is responsible for keeping track of the messages being sent between agents. This would allow us to see the reasons
for some of the emerging behavior of agents.
The experiment was used to test various hypotheses and understand how
learning and behavior are seen in a labor and goods market model. The simulation was started with neither the malls nor the people having knowledge
about their previous actions and the results were as follows.
Best performing malls. The malls were told to go bankrupt if their capitals
went below a bankruptcy level. Therefore, whenever this happened, the
malls were removed from the simulation. The list of functions which the
malls could include in their strategies were