EXemplar-based attention to distinctive InpuT model (Kruschke, 2001)
List of model parameters
R-by-C matrix of training items
The contents of this help file are relatively brief; a more extensive tutorial on using slpEXIT can be found in Spicer et al. (n.d.).
The functions works as a stateful list processor. Specifically, it takes a data frame as an argument, where each row is one trial for the network, and the columns specify the input representation, teaching signals, and other control signals. It returns a matrix where each row is a trial, and the columns are the response probabilities at the output units. It also returns the final state of the network (cue -> exemplar, and cue -> outcome weights), hence its description as a 'stateful' list processor.
References to Equations refer to the equation numbers used in the Appendix of Kruschke (2001).
tr must be a data frame, where each row is one trial
presented to the network, in the order of their occurence.
tr requires the following columns:
x1, x2, ... - columns for each cue (
1 = cue present,
0 = cue absent). These columns have to start with
ascending with features
..., x2, x3, ... at adjacent
columns. See Notes 1, 2.
t1, t2, ... - columns for the teaching values indicating the
category feedback on the current trial. Each category needs a single
teaching signal in a dummy coded fashion, e.g., if the first category
is the correct category for that trial, then
t1 is set to
1, else it is set to
0. These columns have to start with
t1 ascending with categories
..., t2, t3, ... at
ctrl - vector of control codes. Available codes are: 0 = normal
trial, 1 = reset network (i.e. reset connection weights to the values
st). 2 = freeze learning. Control codes are
actioned before the trial processed.
opt1, opt2, ... - optional columns, which may have any name you
wish. These optional columns are ignored by this function, but you may
wish to use them for readability. For example, you might include
columns for block number, trial number, and stimulus ID..
st must be a list containing the following items:
nFeat - integer indicating the total number of possible
stimulus features, i.e. the number of
x1, x2, ... columns in
nCat - integer indicating the total number of possible
categories, i.e. the number of
t1, t2, ... columns in
phi - response scaling constant - Equation (2)
c - specificity parameter. Defines the narrowness of
receptive field in exemplar node activation - Equation (3).
P - Attentional normalization power (attentional capacity) -
Equation (5). If
1 then the attention weights
will satisfy the constraint that attention strength for currently
present features will sum to one. The sum of attention strengths for
present features grows as a function of
l_gain - attentional shift rate - Equation (7)
l_weight - learning rate for feature to category associations.
- Equation (8)
l_ex - learning rate for exemplar_node to gain_node associations
- Equation (9)
iterations - number of iterations of shifting attention on each
trial (see Kruschke, 2001, p. 1400). If you're not sure what to use
here, set it to 10.
sigma - Vector of cue saliences, one for each cue. If you're
not sure what to put here, use 1 for all cues except the bias cue. For
the bias cue, use some value between 0 and 1.
w_in_out - matrix with
nFeat columns and
defining the input-to-category association weights, i.e. how much each
feature is associated to a category (see Equation 1). The
columns follow the same order as
x1, x2, ... in
and likewise, the
nCat rows follow the order of
t1, t2, ....
exemplars - matrix with
nFeat columns and n rows, where
n is the number of exemplars, such that each row represents a single
exemplar in memory, and their corresponding feature values.
nFeat columns follow the same order as
x1, x2, ...
tr. The n-rows follow the same order as in the
w_exemplars matrix defined below. See Note 3.
w_exemplars - matrix which is structurally equivalent to
exemplars. However, the matrix represents the associative weight
from the exemplar nodes to the gain nodes, as given in Equation 4.
nFeat columns follow the same order as
x1, x2, ... in
tr. The n-rows follow the same order
as in the
Returns a list containing three components (if xtdo = FALSE) or four
components (if xtdo = TRUE,
g is also returned):
Matrix of response probabilities for each outcome on each trial
Matrix of final cue -> outcome associative strengths
Matrix of final cue -> exemplar associative strengths
Vector of gains at the end of the final trial
1. Code optimization in slpEXIT means it's essential that every cue is
either set to 1 or to 0. If you use other values, it won't work
properly. If you wish to represent cues of unequal salience, use
2. EXIT simulations normally include a 'bias' cue, i.e. a cue that is
present on all trials. You will need to explicitly include this in
your input representation in
tr. For an example, see the output
3. The bias cue should be included in these exemplar representations,
i.e. they should be the same as the representation of the stimuli in
tr. For an example, see the output of
René Schlegelmilch, Andy Wills, Angus Inkster
Kruschke, J. K. (1996). Base rates in category learning. Journal of Experimental Psychology-Learning Memory and Cognition, 22(1), 3-26.
Kruschke, J. K. (2001). The inverse base rate effect is not explained by eliminative inference. Journal of Experimental Psychology: Learning, Memory & Cognition, 27, 1385-1400.
Spicer, S.G., Schlegelmilch, R., Jones, P.M., Inkster, A.B., Edmunds, C.E.R. & Wills, A.J. (n.d.). Progress in learning theory through distributed collaboration: Concepts, tools, and examples. Manuscript in preparation.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.