function model = config(savePath)
%% Sparse/Efficient Coding
%###############################################################################################################
%parameters in brackets represent those used for multiple scales
Basis_num_used = [10, 10]; % number of basis used to encode in sparse mode
Basis_size = 8.^2*2; % size of each (binocular) base vector
Basis_num= [300,300]; % total basis number
eta = 0.00005; % 0.05; % learning rate for base adaptation
Temperature = 0.01; % temperature in softmax encoding [not used @tm]
Dsratio = [4, 1]; % the distance of n pixel equals 'n/Dsratio' Pixel after downsampling. Usual values: 1,2,4,8
n_scales = length(Basis_num); % number of scales
patch_size = [8, 8]; % size of the extracted patches (patch_size x patch_size)
%-------------
for s = 1:n_scales
PARAMSC{s} = {Basis_num_used(s), Basis_size, Basis_num(s), eta, Temperature, Dsratio(s), n_scales, patch_size(s)};
end
%% Reinforcement Learning
%###############################################################################################################
action = 1:5; % number of actions for RL agents
alpha_v = 0.02; % learning rate to update the value network(0.05)
alpha_n = 0.01; % learning rate of natural policy gradient (0.05)
alpha_p = 0.008; % learning rate to update the policy network (1)
xi = 0.6; % td discount factor
gamma = 0.01; % learning rate to update cumulative value/ mean reward; (0.01)
Temperature = 1; % temperature in softmax function in policy network
S0 = sum(Basis_num); % number of neurons in the input layer
weight_range = [0.35,0.05]; % maximum initial weight
lambda = 0.04; % reguralization factor - additional weight decay term
Act_blur = [-2 -1 0 1 2]; % action spaces of acc RL agent
Act_disp = [-2 -1 0 1 2]; % action spaces of verg RL agent
%-------------
PARAMRL_BLUR = {action,alpha_v,alpha_n,alpha_p,xi,gamma,Temperature,lambda,S0,weight_range,Act_blur};
PARAMRL_DISP = {action,alpha_v,alpha_n,alpha_p,xi,gamma,Temperature,lambda,S0,weight_range,Act_disp};
%% setup parameters for model
%###############################################################################################################
trainTime = 5*10^6; % total number of interations in the training, in total. Usually 5 000 000
Interval = 10; % Number of iterations in one fixation, i.e., before input texture is repositioned
saveAt = 0 : ceil(trainTime / 20) : trainTime; % 20+1 datapoints during training where everything is saved
spectacles_l = 0; % lense in front of left eye which leads lense to focus at different position plane when relaxed
spectacles_r = 0; % lense in front of right eye which leads lense to focus at different position plane when relaxed
cataract_l = 0; % additional blur in left eye. 0: healthy condition, >0: stdv (in px) of additional gaussian blur on left eye
cataract_r = 0; % additional blur in right eye. 0: healthy condition, >0: stdv (in px) of additional gaussian blur on right eye
useSuppression = 1; % toggles suppression on or off
threshold = 0.6; % activation threshold of contrast units
saturation = 0.8; % saturation of contrast units
slope = saturation/(1-threshold); % slope of non-linearity of contrast units
eta_supp = 0.1; % learning rate for exponential runing average
exci = 1; % switches excitatory effect of suppression model on or off
suppr = 1; % switches suppresive effect of suppression model on or off
SUPARAMS = {useSuppression, threshold, slope, eta_supp, exci, suppr};
% more general simulation-related parameters
trainRL = 1; % turns reinforcement learning on or off
trainSC = 1; % turns sparse coding on or off
n_textures = 300; % number of textures used as inputs
normFeat = 1; % indicates if the feature vector should be normalized
max_obj_plane = 3; % target plane positioned between +-max_obj_plane
max_acc_disp_plane = 3; % acc and disp plane between +- max_acc_disp_plane
max_disp = max_obj_plane + max_obj_plane; % largest absolute possible disparity value
max_blur = max_obj_plane + max_acc_disp_plane; % largest possible blur value
% parameters related to input rendering
texture_directory = './textures/'; % folder that contains the input images
image_size = 300; % size of input images
colored_input_images = 0; % use gray values or rgb
window_size = 280; % size of initial cut out window before disparity is applied
blur_scale = 0.8; % set relation between 1 [a.u.] of accommodation error and stdv of defocus blur filter
alpha = 10^-3; % update rate for running averages of reward normalization
PARAMModel = {0,trainTime,Interval, ...
spectacles_l,spectacles_r,cataract_l,cataract_r,savePath,trainRL,max_disp,max_blur,saveAt, ...
SUPARAMS,window_size,alpha,blur_scale,trainSC,n_textures, ...
texture_directory,image_size,colored_input_images,max_obj_plane,max_acc_disp_plane};
%% ###############################################################################################################
PARAM = {PARAMModel,PARAMSC,PARAMRL_BLUR,PARAMRL_DISP};
model = Model(PARAM); %create model instance