demo.m

%% DEMO for Heterogeneous Activation NEAT
% The algorithm and data configuration are loaded, with the data being
% partitioned into k folds. The algorithm is executed and data collected.
% Configuration and results are saved in a Matlab binary (.mat)
% Algorithm can be run in headless mode

% Author: Alexander Hagg
% Bonn-Rhein-Sieg University of Applied Sciences (HBRS)
% email: alexander.hagg@h-brs.de
% Jan 2017; Last revision: 30-Mar-2017

clear; clc; set(0,'DefaultFigureWindowStyle','docked');
algo                        = default_params;
results                     = [];
algo.num_replicates         = 2;
algo.headless               = true;

% Datasets: data_cancer, data_cholesterol, data_engine
algo.data                   = get_folds(data_cancer, 3); % Set number of validation folds


disp(['Dataset: ' algo.data.name]);
for rep = 1:algo.num_replicates
    disp(['=== Replicate ' int2str(rep) '===']);
    for partition = 1:algo.data.num_partitions
        disp(['Training partition ' int2str(partition)]);
        algo.current_run = partition;
        
        [pop, run_data] = run_neat(algo);
        
        results = collect_data(rep,partition,run_data,algo,results);
        save(algo.data.name,'results','algo','-v7.3');
    end
end

%% Visualize all folds of all runs
figure(999)
semilogy(results.test_error','LineWidth',2);
grid on;xlabel('Generations');ylabel('MSE');
axis([0 algo.maxGen 1e-2 1e0]);