Searching for the best set of parameter values is a key component of cognitive modelling and one in which a great deal of uncertainty lies. Parameter search can be a slow, laborious process when done by hand, particularly when a model has several interacting parameters, and can be challenging when models are non-differentiable, non-continuous, nonlinear, stochastic, or have many local optima. There a several methods for searching parameter spaces for such models. Here I present two: differential evolution (DE) and using a High Throughput Computing (HTC) environment managed by HTCondor. The two methods are similar in that they both explore parameter spaces by generating populations of models, but there the similarity ends. Below I describe both, explain the circumstances where choosing one may be preferable over the other, and provide an example of each using a simple ACT-R model for the reader to investigate.
|Title of host publication||Proceedings of ICCM 2016|
|Subtitle of host publication||14th International Conference on Cognitive Modeling|
|Editors||David Reitter, Frank E. Ritter|
|Publisher||The Pennsylvania State University Press|
|Number of pages||2|
|Publication status||Published - 1 Aug 2016|
|Event||14th International Conference on Cognitive Modeling - University Park, Penn State University, Pennsylvania, United States|
Duration: 3 Aug 2016 → 6 Aug 2016
Conference number: 14
|Conference||14th International Conference on Cognitive Modeling|
|Abbreviated title||ICCM 2016|
|Period||3/08/16 → 6/08/16|
Peebles, D. (2016). Two methods for search and optimising cognitive model parameters. In D. Reitter, & F. E. Ritter (Eds.), Proceedings of ICCM 2016: 14th International Conference on Cognitive Modeling (pp. 234-235). The Pennsylvania State University Press.