%==========================================================================
%Title: Phasic dopamine changes and Hebbian mechanisms during probabilistic
%reversal learning in striatal circuits: a computational study
%
%Description: Script to perform the testing procedure: 2 choices
%experiment.
% To be run after 'Basal_Training_synapses_2_channels.m' or Reversal_Training_synapses_2_channels.m’.
% The script calls the function used to simulate the task named
% 'BG_model_Response_Stimuli_Task1'.
%
% 100 trials are executed, by using the inputs S1 and S2 50 times each.
% During the test, random Gaussian noise was applied not only to cortical
% neurons but also to the input stimuli (simulating a noisy environment).
%
% This testing procedure has been performed starting from the network
% obtained after different epochs of learning indicated by vectors in lines
% 46-47.
%
% The results are shown at the end of the simulation in a table organized
% as follow:
% 1st column: epoch
% 2nd column: mean 1st Action
% 3rd column: standard deviation 1st Action
% 4th column: mean 2nd Action
% 5th column: standard deviation 2nd Action
%
% Mauro Ursino, Miriam Schirru Jan. 2022
%==========================================================================
clc
clear
close all
oldpath = path;
selpath = uigetdir;
path(selpath,oldpath)
answer = input('choose the rule (1 = post-post, 2 = post-pre, 3 = pre-pre, 4 = total, 5 = Oja, 6 = adapt )');
answer1 = input ('choose the training (1 = basal, 2 = reversal)');
% %--------------------------------------------------------------------------
% % Initialisation
% %--------------------------------------------------------------------------
Nprove = 10;
Ns = 2;
Nc = 2;
% Epoche = [1 50 75 100 150 200 250 300 350 400]; % case of reversal
Epoche = [1 25 50 75 100 150 200 ]; % case basal
L_epoche = length(Epoche);
Matrice_risultati = zeros(L_epoche,5);
for indice_epoca = 1 : L_epoche
Epoca = Epoche(indice_epoca);
Array_n_succ_tot = zeros(Nc,Nprove);
Array_n_small_succ_tot = zeros(Nc,Nprove);
S1 = zeros(Nc,1);
S2 = zeros(Nc,1);
% S3 = zeros(Nc,1);
% S4 = zeros(Nc,1);
for ii = 1:Nprove
switch answer
case 1
if answer1 == 1
name = strcat('W_tot_post_post_',num2str(ii));
elseif answer1 == 2
name = strcat('Rev_W_tot_post_post_',num2str(ii));
end
case 2
if answer1 == 1
name = strcat('W_tot_post_pre_',num2str(ii));
elseif answer1 == 2
name = strcat('Rev_W_tot_post_pre_',num2str(ii));
end
case 3
if answer1 == 1
name = strcat('W_tot_pre_pre_',num2str(ii));
elseif answer1 == 2
name = strcat('Rev_W_tot_pre_pre_',num2str(ii));
end
case 4
if answer1 == 1
name = strcat('W_tot_total_',num2str(ii));
elseif answer1 == 2
name = strcat('Rev_W_tot_total_',num2str(ii));
end
case 5
if answer1 == 1
name = strcat('W_tot_oja_',num2str(ii));
elseif answer1 == 2
name = strcat('Rev_W_tot_oja_',num2str(ii));
end
case 6
if answer1 == 1
name = strcat('W_tot_post_adapt_',num2str(ii));
elseif answer1 == 2
name = strcat('Rev_W_tot_post_adapt_',num2str(ii));
end
end
load(name)
% basal stimuli
caso = 0;
STN_ON = 1;
T_ON = 1;
%trained synapses
Wgc = squeeze(Wgc_epocs(:,:,2*Epoca));
Wgs = squeeze(Wgs_epocs(:,:,2*Epoca));
Wnc = squeeze(Wnc_epocs(:,:,2*Epoca));
Wns = squeeze(Wns_epocs(:,:,2*Epoca));
if answer1 == 1
S_high = 1.0;
S_small = 0.3;
elseif answer1 == 2
S_high = 0.3;
S_small = 1.0;
end
%S1: stimulus 1
S1(1) = S_high;
S1(2) = S_small;
% S1(3) = 0.1;
% S1(4) = 0.1;
% % S1 = S1';
Correct_winner_1 = 1;
Small_winner_1 = 2;
%S2: stimulus 2
S2(1) = S_small;
S2(2) = S_high;
% S2(3) = 0.1;
% S2(4) = 0.1;
% S2 = S2';
Correct_winner_2 = 2;
Small_winner_2 = 1;
%S3: stimulus 3
S3(1) = 0.1;
S3(2) = 0.1;
S3(3) = S_high;
S3(4) = S_small;
% S3 = S3';
Correct_winner_3 = 3;
Small_winner_3 = 4;
%S4:stimulus 4
S4(1) = 0.1;
S4(2) = 0.1;
S4(3) = S_small;
S4(4) = S_high;
% S4 = S4';
Correct_winner_4 = 4;
Small_winner_4 = 3;
% task parameters
trials = 100/Ns;
Dop_tonic = 1.0; % value of the dopaminergic input used during training, default 1.2
S_vett = [];
exitC = [];
Winner = [];
S_vett_tot = zeros(Nc,Ns*trials);
exitC_tot = zeros(Nc,Ns*trials);
Winner_tot = zeros(1,Ns*trials);
n_err_act_tot = zeros(1,Ns*trials);
rng(21)
noise1=zeros(Nc,Ns*trials);
noise1(1:Ns,1:Ns*trials) = 0.15*randn(Ns,Ns*trials);
% noise1 = 0.20*randn(Nc,4*trials);% noise to the cortex
rng(31)
noise2=zeros(Nc,Ns*trials);
noise2(1:Ns,1:Ns*trials) = 0.2*randn(Ns,Ns*trials);% noise to S
%%
for i = 1:trials
action = randperm (Ns);
for j = 1:Ns
noiseS = noise2(:,j+Ns*(i-1));
noiseC = noise1(:,j+Ns*(i-1));
%
switch action(j)
case 1
S = S1;
S(1) = S(1)+noiseS(1);
S(2) = S(2)+noiseS(2);
% S(3) = S(3)+noiseS(3);
% S(4) = S(4)+noiseS(4);
Correct_winner = Correct_winner_1;
Minor_winner = Small_winner_1;
case 2
S = S2;
S(1) = S(1)+noiseS(1);
S(2) = S(2)+noiseS(2);
% S(3) = S(3)+noiseS(3);
% S(4) = S(4)+noiseS(4);
Correct_winner = Correct_winner_2;
Minor_winner = Small_winner_2;
case 3
S = S3;
S(1) = S(1)+noiseS(1);
S(2) = S(2)+noiseS(2);
S(3) = S(3)+noiseS(3);
S(4) = S(4)+noiseS(4);
Correct_winner = Correct_winner_3;
Minor_winner = Small_winner_3;
case 4
S = S4;
S(1) = S(1)+noiseS(1);
S(2) = S(2)+noiseS(2);
S(3) = S(3)+noiseS(3);
S(4) = S(4)+noiseS(4);
Correct_winner = Correct_winner_4;
Minor_winner = Small_winner_4;
end
%%
S(find(S>1)) = 1;
S(find(S<0)) = 0;
% S(3:4) = 0;
% Call to the function which simulates the basal ganglia response
[Uc,C,Ugo,Go,IGo_DA_Ach,Unogo,NoGo,INoGo_DA_Ach,Ugpe,Gpe,Ugpi,Gpi,Ut,T,Ustn,STN,E,k_tap_vett,Uchi,ChI,t] = BG_model_Response_Stimuli_Task1(S,Wgc,Wgs,Wnc,Wns,STN_ON,T_ON,Dop_tonic,noiseC);
S_vett(1,j) = S(1);
S_vett(2,j) = S(2);
% S_vett(3,j) = S(3);
% S_vett(4,j) = S(4);
Winner(j) = Correct_winner;
Small_winner(j) = Minor_winner;
exitC(1,j) = C(1,end); %exit cortex ch 1
exitC(2,j) = C(2,end);
if max(exitC(:,j)) < 0.9 | sum (exitC(:,j)>0.9)>1 % number of no act or multiple act
n_err_act(j) = 1;
exitC(:,j) = zeros(Nc,1); %multiple act or no act
else
n_err_act(j)= NaN;
end
end
S_vett_tot(:,Ns*i-(Ns-1):Ns*i) = S_vett;
exitC_tot (:,Ns*i-(Ns-1):Ns*i) = exitC;
Winner_tot(1,Ns*i-(Ns-1):Ns*i) = Winner;
Small_winner_tot(1,Ns*i-(Ns-1):Ns*i) = Small_winner;
err = isempty(n_err_act);
if err ==0
n_err_act_tot(1,Ns*i-(Ns-1):Ns*i) = n_err_act; %no act or mult act
end
end
%%
n_succ_tot = zeros(Ns,1);
n_small_succ_tot = zeros(Ns,1);
idx_act_C1 = find(exitC_tot(1,:)>=0.9);
lostC1 = isempty(idx_act_C1);
idx_act_C2 = find(exitC_tot(2,:)>=0.9);
lostC2 = isempty(idx_act_C2);
if lostC1==0
n_succ_tot(1) = length(intersect(find(Winner_tot==1),idx_act_C1));
n_small_succ_tot(1) = length(intersect(find(Small_winner_tot==1),idx_act_C1));
% n_err_tot(1) = length(setdiff(find(Winner_tot==1),idx_act_C1));
end
if lostC2==0
n_succ_tot(2) = length(intersect(find(Winner_tot==2),idx_act_C2));
n_small_succ_tot(2) = length(intersect(find(Small_winner_tot==2),idx_act_C2));
% n_err_tot(2) = length(setdiff(find(Winner_tot==2),idx_act_C2));
end
n_succ = sum(n_succ_tot);
n_small_succ = sum(n_small_succ_tot);
n_succ_tot;
n_small_succ_tot;
Array_n_succ_tot(:,ii) = n_succ_tot;
Array_n_small_succ_tot(:,ii) = n_small_succ_tot;
end
Array_n_succ = sum(Array_n_succ_tot);
Array_n_small_succ = sum(Array_n_small_succ_tot);
Array_error= 100 - Array_n_succ - Array_n_small_succ;
mu_succ = mean(Array_n_succ);
std_succ = std(Array_n_succ);
mu_small_succ = mean(Array_n_small_succ);
std_small_succ = std(Array_n_small_succ);
Matrice_risultati(indice_epoca,:) = [Epoca mu_succ std_succ mu_small_succ std_small_succ];
end
% disp(Matrice_risultati)
TEST=array2table(Matrice_risultati);
TEST.Properties.VariableNames={'Epoch','Mean 1st Action','Std 1st Action','Mean 2nd Action','Std 2nd Action'};
TEST