# PARAMETERS AND OPTIONS
# Each of the following lines consists of the name of a parameter/option, an equal sign, and a number - all separated by a space.
# Please do not change the name of the parameter/option. 
# The denotations, that are included in brackets within the comment sections, refer to those used in Harris & Livesey (2010). 


# ---------------------------------------------------------
# BASIC STRUCTURE OF THE NETWORK AND THE SIMULATION
  
# Number of time steps in each trial. Please make sure that you specify enough time steps to present all stimuli and record 
# all activation as defined in the trial definition. Also bear in mind that stimuli decay gradually, i.e. a stimulus' units 
# may still be active after its offset. 
# We advise allowing at least 25 ticks beyond the offset of the last stimulus.
num_TicksPerTrial = 50

# Number of elements representing each stimulus. Each stimulus specified in the stimulus definition is represented by several 
# elements that in turn consists of one E, I and A unit each. The activation of each of these units and their associations with each 
# other have to be updated in each time step using several iterations. Increasing the number of elements increases computing time 
# substantially.
num_elements = 10
  
# Connectivity between E units. This parameter specifies the likelihood that one E unit is connected to another E unit. The 
# likelihoods of the two unidirectional connections between two E units are independent. Increasing the likelihood increases
# the number of connections between E units and thus the learning speed.  A likelihood of .5 maximises the variability across the
# network that is necessary to solve certain learning tasks.If p is <1, the output of any single simulation will vary, and it is
# therefore advisable to average across repeated simulations. For many simulated experiments, the connectivity does not have important 
# consequences for the simulation, and thus we advise setting p = 1 to avoid variability between simulations.
connect_p = 1
 
 
# ---------------------------------------------------------
# PARAMETERS FOR ACTVATION FUNCTION

# Rule to calculate external sensory input (Sx) by a stimulus to its elements. Each stimulus activates a set of elements (or more 
# precisely, its E and A units). Their activation is determined by the minimal and maximal activation of this stimulus as specified
# in the stimulus definition and the rule selected here. The simulator allows the user to choose between three different rules to 
# determine this external sensory input for each element: 
# 1 = linearly:  input strengths are graded across elements of each stimulus in equal steps from the maximal to the minimal activation 
# 2 = randomly: for each element a value is drawn from the uniform distribution on the activation interval [minimum, maximum]
# 3 = constant: all elements are activated to the mean of the minimal and maximal activation
# If option 2 is selected, the output of any single simulation will vary, and it is therefore advisable to average across repeated 
# simulations. For many simulated experiments, the there is little difference between options 1 and 2 for the simulation. In such cases,
# we advise selecting option 1 to avoid variability between simulations.
S_distribution = 1

# Parameter for response potential functions of E, I and A units (Rpot):
# Exponent (p) of response potential function (Rpot). Higher values of p make the response potential functions sigmoid and increase the
# maximal slope.
Power_E = 2
Power_I = 2
Power_A = 5
# Constant (D) of response potential function (Rpot). This parameter affects the relative sensitivity of the response potential to
# an input and prevents the denominator ever being zero
D = 0.04

# Rate parameters (delta) for change of the activation function (dR/dt). These two parameters control how fast the activation of
# E, I, and A units actually rises and decays, respectively.
# Rate parameters (delta) for rising of the activation (R), i.e. when the unit's response potential to its current input is higher
# than its current activation ((Rpot -R)>0)
Growth_E = 0.5
Growth_I = 0.5
Growth_A = 0.5
# Rate parameters (delta) for decay of the activation (R), i.e. when the unit's response potential to its current input is lower than
# its current activation ((Rpot -R)<0)
Decay_E = 0.2
Decay_I = 0.2
Decay_A = 0.2


# ---------------------------------------------------------
# FIXED CONNECTION WEIGHTS AND NORMALIZATION

# Normalisation based on similarity: Connection weight from E to I units of different elements (z). E units activate other elements' 
# I units and thus increase the inhibition of E units by I units, i.e. normalisation in the network. Instead of a graded change in
# values of z across elements, three different degrees of normalisation can be specified: 
# (1) Normalisation of I units by E units of elements belonging to other stimuli.
norm_across_all = 0.01
# (2) Normalisation of I units by E units of elements of other stimuli that share the same modality. The modality is specified in
# the stimulus definition file. Stimuli with a modality of 0 do not share modality with any other stimulus. The follwoing value
# will be divided number of elements in each stimulus.
norm_within_modality = 0.5 
# (3) Normalisation of I units by E units of other elements of the same stimulus. This value will be divided by the number of elements
# in the stimulus.
norm_within_stim = 0.5

# Normalisation based on A units
# Connection weight from A to I unit (ka). A units inhibit I units of the same elements. Increasing the connection weight therefore
# decreases normalisation of the E units. The parameter value should be positive (even though the connection is inhibitory). 
Weight_IA = 4
# Connection weight between A units (w): A units are in turn inhibited by other elements' A units. Increasing this connection weight
# therefore increases normalisation of E units. The parameter value should be positive (even though the connection is inhibitory). 
Weight_AA = 0.1
 
 
# ---------------------------------------------------------
# LEARNING PARAMETER 

# Parameter for coactivation function of two elements (DELTA). Changes in the connection weights between E units (i.e. the associative
# strength V) depend on the coactivation of the two elements. The coactivation changes in response to increments in the recipient's E
# and I units and is scaled by two rate parameters, betaE and betaI. As default, both parameters are set to 0 when the activation of E and I
# is decaying, so that associative changes are driven only by increases in elements' activation. Increasing the parameter for rising
# activation accelerates learning. 
# Parameter (betaE) for the effects of rising of recipient’s E unit's activation (dE_dt>0)
Beta_E_pos = 0.02
# Parameter (betaE) for the effects of decay of recipient's E unit's activation (dE_dt<0)
Beta_E_neg = 0.0
# Parameter (betaI) for the effects of rising of recipient’s I unit's activation (dI_dt>0)
Beta_I_pos = 0.1
# Parameter (betaI) for the effects of decay of recipient’s I unit's activation (dI_dt<0)
Beta_I_neg = 0.0

# Rate parameter (kV) for changes in connections weights function (dV_dt). This parameter controls how fast associative weights
# are actually changing. The parameter value should be between 0 and 1. High values accelerates learning; low values results
# in more stable learning.
k_V = 1

# Rate parameter (kalpha) for changes in associability of sender's E unit function (dalpha_dt). This parameter controls how fast
# associability is actually changing. The parameter should have values between 0 and 1.
k_Alpha = 1


# ---------------------------------------------------------
# OUTPUT OPTIONS

# Figures: 
# Which figures should be created? 0 = no; 1= yes:
# Final associative weights (i.e. connections between E units) at the end of the experiment
fig_V = 1
# Activation of all E units in each time step
fig_E = 0
# Activation of all I units in each time step
fig_I = 0
# Activation of all A units in each time step
fig_A = 0
# E units' activation, recorded and averaged for each trial as specified in the trial definition
fig_CR = 1

# Output: how should the E units' activation, recorded and averaged for each trial as specified in the trial definition, be saved?
# 0= not at all, 
# 1= as xls: each phase is saved in a separate sheet with the corresponding name
# 2= as Matlab mat-file: each phase is saved in a cell of a cell array
mat_CR = 1


#------------------------------------------------
# lines beginning with a hash are ignored
# This file is part of a simulation programm for the Attention-Modulated Assocative Network (AMAN) 
# that can be downloaded from http://sydney.edu.au/science/psychology/staff/justinh/downloads/  
# (see also Thorwart, Livesey & Harris, 2010)