shin92exalcove_opt: Parameter optimization of ex-ALCOVE model with shin92 CIRP

In catlearn: Formal Psychological Models of Categorization and Learning

Parameter optimization of ex-ALCOVE model with shin92 CIRP

Description

Uses shin92exalcove to find best-fitting parameters for the ex-ALCOVE model for the shin92 CIRP.

Usage

  shin92exalcove_opt(params = c(2, 1, 0.25, 0.75), recompute = FALSE,
  trace = 0)

Arguments

params	A vector containing the initial values for c, phi, la, and lw, in that order. See slpALCOVE for an explanation of these parameters. Where recompute is FALSE, this argument has no effect.
recompute	When set to TRUE, the function re-runs the optimization (which takes about 25 minutes on a 2.4 GHz processor). When set to FALSE, the function returns a stored copy of the results of the optimization (which is instantaneous).
trace	Sets the level of tracing information (i.e. information about the progress of the optimization), as defined by the optim function. Set to 6 for maximally verbose output. Where recompute is FALSE, this argument has no effect.

Details

This function is an archive of the optimization procedure used to derive the best-fitting parameters for the shin92exalcove simulation; see Spicer et al. (2017) for a tutorial introduction to the concept of simulation archives.

Optimization used the L-BFGS-B method from the optim function of the standard R stats package. The objective function was sum of squared errors. Please inspect the source code for further details (e.g. type shin92exalcove_opt).

This function was run in 16 times from different starting points, using 8 threads on a Core i7 3.6 GHz processor. The default parameters of this function are those for the best fit from those 16 starting points. The 16 starting points were

pset <- rbind( c(2,1,.25,.25),c(2,1,.25,.75),c(2,1,.75,.25),c(2,1,.75,.75), c(2,3,.25,.25),c(2,3,.25,.05),c(2,3,.75,.25),c(2,3,.75,.75), c(8,1,.25,.25),c(8,1,.25,.75),c(8,1,.75,.25),c(8,1,.75,.75), c(8,3,.25,.25),c(8,3,.25,.75),c(8,3,.75,.25),c(8,3,.75,.75) )

not all of which converged successfully.

Value

A vector containing the best-fitting values for c, phi, la, and lw, in that order. See slpALCOVE for an explanation of these parameters.

Author(s)

Andy Wills

References

Spicer, S., Jones, P.M., Inkster, A.B., Edmunds, C.E.R. & Wills, A.J. (2017). Progress in learning theory through distributed collaboration: Concepts, tools, and examples. Manuscript in preparation.

catlearn documentation built on April 4, 2023, 5:12 p.m.