MetaController/learn.m at master · DevComPsy/MetaController · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
function [mc] = learn(mc,task,settings,g)
%function simulating the Q-learning of each system

%% calc PEs for each system

for i = 1:settings.nsystems

    %% get chosen Q value of system
    q_val_chosen = [];
    if mc.dims(i) == 1
        q_val_chosen = mc.Q{i}(task.chosenstimulus(1,g),end);
    elseif mc.dims(i) == 2
        q_val_chosen = mc.Q{i}(task.chosenstimulus(1,g),task.chosenstimulus(2,g),end);
    elseif mc.dims(i) == 3
        q_val_chosen = mc.Q{i}(task.chosenstimulus(1,g),task.chosenstimulus(2,g),task.chosenstimulus(3,g),end);
    else
        error('undefined model')
    end

    mc.pe{i}(g) = task.outcome(g) - q_val_chosen;

    %% update Q value for chosen state
    if mc.dims(i) == 1
        mc.Q{i}(:,end+1) = mc.Q{i}(:,end);
        mc.Q{i}(task.chosenstimulus(1,g),end) =  mc.Q{i}(task.chosenstimulus(1,g),end) + settings.alpha * mc.pe{i}(g);
    elseif mc.dims(i) == 2
        mc.Q{i}(:,:,end+1) = mc.Q{i}(:,:,end);
        mc.Q{i}(task.chosenstimulus(1,g), task.chosenstimulus(2,g),end) =  mc.Q{i}(task.chosenstimulus(1,g), task.chosenstimulus(2,g),end) + settings.alpha * mc.pe{i}(g);
    elseif mc.dims(i) == 3
        mc.Q{i}(:,:,:,end+1) = mc.Q{i}(:,:,:,end);
        mc.Q{i}(task.chosenstimulus(1,g), task.chosenstimulus(2,g),task.chosenstimulus(3,g), end) =  mc.Q{i}(task.chosenstimulus(1,g), task.chosenstimulus(2,g), task.chosenstimulus(3,g), end) + settings.alpha * mc.pe{i}(g);
    else
        error('undefined model')
    end

end