doubleint_demo.m

approximate reinforcement learning

% DOUBLEINT_DEMO Demonstration of the use of the approximate RL package, on the double
% integrator problem

% TODO This is a quick-and-dirty built file, should be fine-tuned and made more detailed

echo on

% First, you need to define the problem. To do that, start from the template sample_problem.
% You need to define the dynamics and return function in a separate function; start from
% the template sample_mdp. You can also define a custom plot function, starting from
% sample_plot.
% See doubleint_problem and doubleint_mdp for an example.
pause;

% Here is how to run fuzzy Q-iteration for the double integrator
% You need to specify the problem function and it's recommended that you specify a data file for saving
% the results
% stats = fuzzyqi('run problem=doubleint_problem datafile=demodata');
pause;
stats = fuzzyqi('run problem=doubleint_problem datafile=demodata');
% Statistics of fuzzy Q-iteration are returned by the function
pause;

% Here's how to make a controlled trajectory with the learnt policy
% hist = fuzzyqi('replay problem=doubleint_problem datafile=demodata x0=[-1,0]');
pause;
hist = fuzzyqi('replay problem=doubleint_problem datafile=demodata x0=[-1,0]');
% A history of the controlled trajectory is returned by the function
pause; commandwindow; pause;

% equivalently, all the parameters can be specified in a structure
clear cfg;
cfg.replay = 1;
cfg.problem = 'doubleint_problem';
cfg.datafile = 'demodata';
cfg.x0 = [-1 0]';
pause;
hist = fuzzyqi(cfg);
pause; commandwindow; pause;


% Other useful parameters:
cfg.loadapprox = 'demodata';        % to load precomputed dynamics, rewards 
                                    % and possibly membership degrees from this file
cfg.interph = 0;                    % to disable policy interpolation in replay
cfg.verb = 0;                       % set verbosity 0 to display only the most important messages
cfg.silent = 1;                     % or disable text and graphical output entirely
pause;

% run again with the new config
stats = fuzzyqi(cfg);
pause;
% and replay
hist = fuzzyqi(cfg);
% notice nothing was plotted since silent=1; we plot the history now
pause;
figh = plothistory(hist); 
pause; close(figh);

% for the entire list of configuration parameters, 
% edit fuzzyqi and check the default configuration comments

% for how to specify various types of parameters 
% on a single configuration string, help str2cfg
pause;

close all; echo off; return;