Auto-Calibration for Agent-Based Purchase Modeling Systems
In one embodiment, the present invention is a method for optimizing a parameter in an agent-based model including running an agent-based simulation based upon a set of parameters to provide an outcome. The method further includes determining an error of the outcome, and automatically determining a revised value for one of the parameters of the set of parameters, which revised value is selected to reduce error attributed to the one of the parameters. The method also includes re-running the simulation utilizing the revised value of the one of the parameters.
1 . A method for optimizing a parameter in an agent-based model comprising:
running an agent-based simulation based upon a set of parameters to provide an outcome;
determining an error of the outcome;
automatically determining a revised value for one of said parameters of said set of parameters, which revised value is selected to reduce error attributed to the one of the parameters; and
re-running the simulation utilizing the revised value of the one of said parameters.
2 . The method of claim 1 wherein the second determining step includes determining the revised value of the one of said parameters using a slope intercept method.
3 . The method of claim 2 wherein the slope intercept method includes estimating a slope of an error curve associated with the one of said parameters at a particular point of the error curve, determining where the slope intersects an axis, and using the value of the one of said parameters at that intersection point in the re-running step.
4 . The method of claim 2 wherein the step of estimating the slope of the error curve includes utilizing a modified version of Newton's Method.
5 . The method of 1 wherein the second determining step includes running another agent-based simulation in which other parameters besides the one of said parameters are generally unchanged compared to the original simulation while the value for the one of said parameters is varied compared to the simulation, and wherein the error outcome of the another agent-based simulation is compared to the error of the outcome of the running step.
6 . The method of claim 5 wherein the second determining step includes calculating a first difference between the error of the outcome of the another agent-based simulation and the error of the outcome of the running step, calculating a second difference between the value of the parameter utilized in the another agent-based simulation and the value of the parameter utilized in the running step, and comparing or dividing the first and second differences.
7 . The method of claim 1 wherein the error of the outcome is designated g(P 1 ), and wherein the second determining step includes running another agent-based simulation in which at least some of the parameters besides the one of said parameters are maintained at their previous values, while the value for the one of said parameters is varied by a difference designated ΔP, and wherein the another simulation results in an error outcome value designated g(P 1 +ΔP), and wherein the method includes determining a slope of an error curve, designated g′(P 1 ), by the equation g′(P 1 )≈(g(P 1 +ΔP)−g(P 1 ))/ΔP.
8 . The method of claim 7 wherein the value of the one of said parameters during the simulation is designated P 1 , the value for the revised value of the one of said parameters is designed P 2 , and wherein the value for P 2 is determined by equation P 2 =P 1 −g(P 1 )/g′(P 1 ).
9 . The method of claim 1 wherein the second determining step includes selecting a revised value for the one of said parameters which seeks to minimize error attributed to the one of said parameters.
10 . The method of claim 9 further comprising the step of utilizing a fraction of the difference between an original value and the revised value in the re-running step.
11 . The method of claim 9 wherein the one of said parameters is an attribute of the agent which at least partially determines the agent's actions or reactions during the simulation.
12 . The method of claim 1 wherein the first determining step includes comparing the outcome of the simulation to known historical values.
13 . The method of claim 1 wherein the first determining step includes comparing an attribute of the agents to a desired attribute.
14 . The method of claim 1 further comprising the steps of automatically repeating the first and second determining steps, and the re-running step, until an optimized value for the one of said parameters is determined.
15 . The method of claim 14 wherein the optimized value is a value which minimizes error attributed to the associated parameter.
16 . The method of claim 1 further comprising the step of determining an error of an outcome of the re-run simulation, and comparing the error of the re-run simulation to the error of outcome of the simulation.
17 . The method of claim 1 wherein the second determining step takes into account the absolute value of the error.
18 . The method of claim 1 wherein the agent-based simulation utilizes a plurality of agents, each agent having a plurality of attributes assigned thereto, and wherein the running and re-running steps include introducing each agent to a stimuli and providing the reaction of each agent to the introduced stimuli.
19 . The method of claim 1 wherein the second determining step is repeated for each parameter of the set of parameters.
20 . The method of claim 1 wherein the running and re-running steps are agent-based simulations which predict consumer responses to advertising or promotional efforts.
21 . The method of claim 20 wherein the one of said parameters relates to at least one of sex, age, race, income, television watching habits, Internet usage habits, radio listening habits, price sensitivity, purchase history, purchase occasions or quality sensitivity of an agent.
22 . The method of claim 1 wherein the method is implemented on a processor which carries out the miming step, both determining steps, and the re-running step.
23 . A method for optimizing a parameter in an agent-based model comprising:
running an agent-based simulation based upon a set of parameters to provide an outcome;
determining an error of the outcome;
automatically determining a revised value for one of said parameters of said set of parameters using a slope intercept method; and
re-running the simulation utilizing the revised value of the one of said parameters.
24 . A method for optimizing a parameter in an agent-based model comprising:
running an agent-based simulation based upon a set of parameters to provide an outcome;
automatically determining a revised value for one of said parameters of said set of parameters, which revised value is selected to reduce error attributed to the one of the parameters; and
re-running the simulation utilizing the revised value of the one of said parameters.