layer {leabRa}

Leabra layer class

Description

This class simulates a biologically realistic layer of neurons in the Leabra framework. It consists of several unit objects in the variable (field) units and some layer-specific variables.

Usage

layer

Format

R6Class object

Value

Object of R6Class with methods for calculating changes of activation in a layer of neurons.

Fields

units

A list with all unit objects of the layer.

avg_act

The average activation of all units in the layer (this is an active binding).

n

Number of units in layer.

weights

A receiving x sending weight matrix, where the receiving units (rows) has the current weight values for the sending units (columns). The weights will be set by the network object, because they depend on the connection to other layers.

ce_weights

Sigmoidal contrast-enhanced version of the weight matrix weights. These weights will be set by the network object.

layer_number

Layer number in network (this is 1 if you create a layer on your own, without the network class).

Methods

new(dim, g_i_gain = 2)

Creates an object of this class with default parameters.

dim

A pair of numbers giving the dimensions (rows and columns) of the layer.

g_i_gain

Gain factor for inhibitory conductance, if you want less activation in a layer, set this higher.

get_unit_acts()

Returns a vector with the activations of all units of a layer.

get_unit_scaled_acts()

Returns a vector with the scaled activations of all units of a layer. Scaling is done with recip_avg_act_n, a reciprocal function of the number of active units.

cycle(intern_input, ext_input)

Iterates one time step with layer object.

intern_input

Vector with inputs from all other layers. Each input has already been scaled by a reciprocal function of the number of active units (recip_avg_act_n) of the sending layer and by the connection strength between the receiving and sending layer. The weight matrix ce_weights is multiplied with this input vector to get the excitatory conductance for each unit in the layer.

ext_input

Vector with inputs not coming from another layer, with length equal to the number of units in this layer. If empty (NULL), no external inputs are processed. If the external inputs are not clamped, this is actually an excitatory conductance value, which is added to the conductance produced by the internal input and weight matrix.

clamp_cycle(activations)

Iterates one time step with layer object with clamped activations, meaning that activations are instantaneously set without time integration.

activations

Activations you want to clamp to the units in the layer.

get_unit_act_avgs()

Returns a list with the short, medium and long term activation averages of all units in the layer as vectors. The super short term average is not returned, and the long term average is not updated before being returned (this is done in the function chg_wt() with the methodupdt_unit_avg_l). These averages are used by the network class to calculate weight changes.

updt_unit_avg_l()

Updates the long-term average (avg_l) of all units in the layer, usually done after a plus phase.

updt_recip_avg_act_n()

Updates the avg_act_inert and recip_avg_act_n variables, these variables update before the weights are changed instead of cycle by cycle. This version of the function assumes full connectivity between layers.

reset(random = FALSE)

Sets the activation and activation averages of all units to 0. Used to begin trials from a stationary point.

random

Logical variable, if TRUE the activations are set randomly between .05 and .95 for every unit instead of 0.

set_ce_weights()

Sets contrast enhanced weight values.

get_unit_vars(show_dynamics = TRUE, show_constants = FALSE)

Returns a data frame with the current state of all unit variables in the layer. Every row is a unit. You can choose whether you want dynamic values and / or constant values. This might be useful if you want to analyze what happens in units of a layer, which would otherwise not be possible, because most of the variables (fields) are private in the unit class.

show_dynamics

Should dynamic values be shown? Default is TRUE.

show_constants

Should constant values be shown? Default is FALSE.

get_layer_vars(show_dynamics = TRUE, show_constants = FALSE)

Returns a data frame with 1 row with the current state of the variables in the layer. You can choose whether you want dynamic values and / or constant values. This might be useful if you want to analyze what happens in a layer, which would otherwise not be possible, because some of the variables (fields) are private in the layer class.

show_dynamics

Should dynamic values be shown? Default is TRUE.

show_constants

Should constant values be shown? Default is FALSE.

References

O'Reilly, R. C., Munakata, Y., Frank, M. J., Hazy, T. E., and Contributors (2016). Computational Cognitive Neuroscience. Wiki Book, 3rd (partial) Edition. URL: http://ccnbook.colorado.edu

Have also a look at https://grey.colorado.edu/emergent/index.php/Leabra (especially the link to the 'MATLAB' code) and https://en.wikipedia.org/wiki/Leabra

Examples

l <- layer$new(c(5, 5)) # create a 5 x 5 layer with default leabra values

l$g_e_avg # private values cannot be accessed
# if you want to see alle variables, you need to use the function
l$get_layer_vars(show_dynamics = TRUE, show_constants = TRUE)
# if you want to see a summary of all units without constant values
l$get_unit_vars(show_dynamics = TRUE, show_constants = FALSE)

# let us clamp the activation of the 25 units to some random values between
# 0.05 and 0.95
l <- layer$new(c(5, 5))
activations <- runif(25, 0.05, .95)
l$avg_act
l$clamp_cycle(activations)
l$avg_act
# what happened to the unit activations?
l$get_unit_acts()
# compare with activations
activations
# scaled activations are scaled by the average activation of the layer and
# should be smaller
l$get_unit_scaled_acts()