estimateWeights {ndl}

Estimation of the association weights using the equilibrium equations of Danks (2003) for the Rescorla-Wagner equations.

Description

A function to estimate the weights (associative strengths) for cue-outcome pairs when learning is in equilibrium, using the equilibrium equations for the Rescorla-Wagner model of Danks (2003).

Usage

estimateWeights(cuesOutcomes, removeDuplicates=TRUE, saveCounts=FALSE,
verbose=FALSE, trueCondProb=TRUE, addBackground=FALSE, hasUnicode=FALSE, ...)

Arguments

cuesOutcomes

A data frame with three variables specifying frequency, cues, and outcomes, that may be created with ndlCuesOutcomes or with the accessory script in the inst/scripts directory:

Cues

A character vector specifying the cues. When there is more than one cue, the cues should be separated by underscores.

Outcomes

A character vector specifying the outcomes. When there is more than one outcome, the outcomes should be separated by underscores.

Frequency

A numeric vector specifying the frequency with which a combination of cues and outcomes occurs.

removeDuplicates

A logical specifying whether multiple occurrences of a Cue in conjunction with an individual instance of an Outcome shall each be counted as a distinct occurrence of that Cue (FALSE: default), or only as a single occurrence (TRUE).

saveCounts

A logical specifying whether the co-occurrence matrices should be saved. If set equal to TRUE, the files coocCues.rda and coocCuesOutcomes.rda will be saved in the current working directory.

verbose

If set to TRUE, display diagnostic messages.

addBackground

If you would like to add a background rate for all your cues and outcomes, but did not include an general environment cue to all your events, one will be added for you to the matrices, as specified in Danks (2003). If changed from the default (FALSE) to TRUE, background cues will be added. The name used for the background rates is "Environ", and will be included in the output weight matrix.

trueCondProb

The conditional probability calculations used will be those specified in Danks (2003). If changed from the default (TRUE) to FALSE, the normalization specified in Baayen, et al (2011) is used.

hasUnicode

A logical specifying whether to apply a UTF-8 to integer conversion to the names of the cues. This was implemented to solve issues with differences Unicode cue names.

...

Control arguments to be passed along from ndlClassify and/or ndlCrossvalidate.

Details

Using Rcpp, a C++ based implementation processes all of the data in RAM. The module will check the amount of RAM you have available in your system and warn you if the amount of RAM is insufficient to build your model.

For examples of how the cuesOutcomes data frame should be structured, see the data sets danks, plurals, and serbian. N.B. Empty Cues or Outcomes (effectively having length = 0), e.g. Cues or Outcomes strings with an initial or final underscore or two immediately adjacent underscores, will result in an error.

Value

A matrix with cue-to-outcome association strengths. Rows are cues, and columns are outcomes. Rows and columns are labeled. If addBackground=T, a row named "Environ" will be added to the output.

Acknowledgements

The assistance of Uwe Ligges in getting the C function cooc to work within the R framework is greatly appreciated. This C function was removed in version 0.2.0 and replaced with the C++ function by Cyrus Shaoul.

Note

Add a note here.

Author(s)

Cyrus Shaoul, R. H. Baayen and Petar Milin, with contributions from Antti Arppe and Peter Hendrix.

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M. (2011), An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

estimateActivations, ndlCuesOutcomes, danks, plurals, serbian

Examples

  data(danks)
  estimateWeights(cuesOutcomes=danks)

  data(plurals)
  plurals$Cues <- orthoCoding(plurals$WordForm, grams=1)
  round(estimateWeights(cuesOutcomes=plurals),2)
  
  data(serbian)
  serbian$Cues <- orthoCoding(serbian$WordForm, grams=2)
  serbian$Outcomes <- serbian$LemmaCase
  sw <- estimateWeights(cuesOutcomes=serbian)
  round(sw[1:5,1:6],2)