Skip to contents

xslControl S3 class

Source: R/xslControl-class.R
xslControl-class.Rd

Control arguments for xsl_run()

Usage

xslControl(
  reps = 1,
  start_matrix = NULL,
  test_noise = 0,
  n_sim = 500,
  verbose = FALSE,
  keep_traj = FALSE,
  keep_sims = FALSE,
  objective = c("sse", "binomial", "multinomial"),
  lapse = 0.01
)

new_xslControl(x = list())

Arguments

reps

Number of times to repeat training

start_matrix

Starting matrix

test_noise

Test noise

n_sim

Number of simulations for stochastic models

verbose

Verbosity

keep_traj

Whether each model run should record its per-trial association matrix (xslFit$traj). Off by default: the trajectory is not used by any function in the package, and for a long corpus it is a serious memory cost (one word-by-object matrix per trial, per simulation). Set TRUE only when you want to inspect the learning trajectory yourself.

keep_sims

Whether xsl_run() should retain the full list of per-simulation xslFit objects (fits[[i]]$sims). Off by default; fits[[i]]$responses (an n_sim x n-words matrix of each simulated participant's final per-word accuracy) is always returned instead, at a fraction of the memory.

objective

What xsl_fit() minimizes, and what xsl_run() scores:

  • "sse" (the default): the n_subj-weighted sum of squared errors between model and human accuracy per test item.

  • "binomial": the negative log-likelihood of how many participants got each test item right, given the model's probability of a correct choice. Counts come from the condition's response_matrix, or otherwise from accuracy and n_subj.

  • "multinomial": the negative log-likelihood of which object each participant chose on each test item, given the model's full choice distribution over the item's alternatives (see choice_probabilities()). Needs a response_matrix.

The likelihoods treat every response as an independent draw from the model's prediction, ignoring individual differences, so they overstate the evidence for one model over another; rankings are more trustworthy than the absolute size of AIC or likelihood-ratio differences.

lapse

For the likelihood objectives: the probability that a response is a uniform guess among the item's alternatives rather than the model's choice. Keeps the likelihood finite when a model is certain and a participant chooses otherwise. Fit it by naming "lapse" in xsl_fit()'s free. Ignored by "sse".

x

List with elements train, test, accuracy, n_subj, label, condition

Value

An object of class xslControl