Simulation of CIRP nosof94 with the SUSTAIN model
Runs a simulation of the nosof94 CIRP using the
slpSUSTAIN model implementation and
nosof94train as the input representation.
nosof94sustain(params = c(9.01245, 1.252233, 16.924073, 0.092327))
params |
A vector containing values for r, beta, d, and eta, in
that order, e.g. params = c(8.1, 1.5, 9.71, 0.8). See
|
NOTE: The underlying slpSUSTAIN function is currently written in R, and hence this simulation will take several minutes to run. slpSUSTAIN may be converted to C++ in a future release, which will reduce the run time of this simulation to a few seconds.
A simulation using slpSUSTAIN and
nosof94train, i.e. a simulation of Nosofsky et al. (1994)
with the Love et al. (2004) SUTAIN model.
Other parameters of slpSUSTAIN are set as follows: tau = 0,
initial lambda = 1, initial w = 0, inital cluster
centered on the first stimulus presented to the siumulated
subject. These values are conventions of modelling with SUSTAIN, and
should not be considered as free parameters. They are set within the
nosof94sustain function, and hence can't be changed without
re-writing the function.
The simulation uses 100 simulated subjects. Like the simulations
nosof94exalcove and nosof94protoalcove, all simulated
participants complete 16 blocks of training. This differs from the
Nosofsky et al. (1994) experiment, in which participants are trained to
a criterion of four consecutive errorless 8-trial subblocks.
The simulation by Gureckis (2014) builds this criterion-based training
into their simulation by using a random number generator to turn the
response probability on each trial into a correct or incorrect
response. This feature of the Gureckis (2014) simulation is not
incorporated here, because the instability in ouput this generates
makes parameter optimization (e.g. via optim) less reliable.
A comparison of 10,000 simulated participants in the Gureckis (2014) simulation with 1,000 simulated participants in the current simulation reveals a mean difference in the 96 reported response probabilities of less than 0.01.
A matrix of predicted response probabilities, in the same order and
format as the observed data contained in nosof94.
Lenard Dome, Andy Wills
Love, B. C., Medin, D. L., & Gureckis, T. M. (2004). SUSTAIN: a network model of category learning. Psychological Review, 111, 309-332.
Gureckis, T. M. (2014). sustain_python. https://github.com/NYUCCL/sustain_python
Nosofsky, R.M., Gluck, M.A., Plameri, T.J., McKinley, S.C. and Glauthier, P. (1994). Comparing models of rule-based classification learning: A replication and extension of Shepaard, Hovland, and Jenkins (1961). Memory and Cognition, 22, 352–369.
Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.