Q-Learning Using Matlab

by Kardi Teknomo

<Contents | Previous | Next >

I have made simple Matlab Code below for this tutorial example and you can modify it for your need. You can copy and paste the two functions into separate text files and run it as ReinforcementLearning.

To model the environment you need to make the instant reward matrix R. Put zero for any door that is not directly to the goal and put value 100 to the door that lead directly to the goal. For unconnected states, use minus Infinity (-Inf) so that it become very negative number. We want to maximize the Q values, thus very negative number will not be considered at all.

The state is numbered 1 to N (in our previous example N = 6). The result of the code is only normalized Q matrix.

You may experiment in the effect of parameter gamma to see how it influences the results.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Q learning of single agent move in N rooms 
% Matlab Code companion of 
% Q Learning by Example, by Kardi Teknomo 
% (http://people.revoledu.com/kardi/)
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
function q=ReinforcementLearning
clc;
format short
format compact

% Two input: R and gamma
% immediate reward matrix; 
% row and column = states; -Inf = no door between room
R=[-inf,-inf,-inf,-inf,   0, -inf;
   -inf,-inf,-inf,   0,-inf, 100;
   -inf,-inf,-inf,   0,-inf, -inf;
   -inf,   0,   0,-inf,   0, -inf;
      0,-inf,-inf,   0,-inf, 100;
   -inf,   0,-inf,-inf,   0, 100];

gamma=0.80;            % learning parameter

q=zeros(size(R));      % initialize Q as zero
q1=ones(size(R))*inf;  % initialize previous Q as big number
count=0;               % counter

for episode=0:50000
   % random initial state
   y=randperm(size(R,1));
   state=y(1);
   
   % select any action from this state
   x=find(R(state,:)>=0);        % find possible action of this state
   if size(x,1)>0,
      x1=RandomPermutation(x);   % randomize the possible action
      x1=x1(1);                  % select an action 
   end

   qMax=max(q,[],2);
   q(state,x1)= R(state,x1)+gamma*qMax(x1);   % get max of all actions 
   state=x1;
   
   % break if convergence: small deviation on q for 1000 consecutive
   if sum(sum(abs(q1-q)))<0.0001 & sum(sum(q >0))
      if count>1000,
         episode        % report last episode
         break          % for
      else
         count=count+1; % set counter if deviation of q is small
      end
   else
      q1=q;
      count=0; % reset counter when deviation of q from previous q is large
   end
end 

%normalize q
g=max(max(q));
if g>0, 
   q=100*q/g;
end
  

The code above is using basic library RandomPermutation below

 
 function y=RandomPermutation(A)
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 % return random permutation of matrix A
 % unlike randperm(n) that give permutation of integer 1:n only,
 % RandomPermutation rearrange member of matrix A randomly
 % This function is useful for MonteCarlo Simulation, 
 %  Bootstrap sampling, game, etc.
 % 
 % Copyright Kardi Teknomo(c) 2005
 % (http://people.revoledu.com/kardi/)
 %
 % example: A = [ 2, 1, 5, 3]
 % RandomPermutation(A) may produce [ 1, 5, 3, 2] or [ 5, 3, 2, 3]
 % 
 % example: 
 % A=magic(3)
 % RandomPermutation(A)
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
   [r,c]=size(A);
   b=reshape(A,r*c,1);       % convert to column vector
   x=randperm(r*c);          % make integer permutation of similar array as key
   w=[b,x'];                 % combine matrix and key
   d=sortrows(w,2);          % sort according to key
   y=reshape(d(:,1),r,c);    % return back the matrix
         

<Contents | Previous | Next >

This tutorial is copyrighted.

Preferable reference for this tutorial is

Teknomo, Kardi. 2005. Q-Learning by Examples. http://people.revoledu.com/kardi/tutorial/ReinforcementLearning/index.html