svm_audio.m

%% Notes
%% Toutes les routines qui sont appelées dans ce programme viennent du cours de 
%% Andrew Ng et peuvent avoir été adaptées. 
%%
%% Les indices commencent a 1
%% Les indices matriciels sont 1 -> row et 2 -> col
%% Pour lire fichier .mat en Octave: load MyMatFile.mat right at the Octave prompt

clear ; close all; clc

set_genre1 = dlmread ('audio-features/party_audio.csv',',',1,0);
set_genre2 = dlmread ('audio-features/dinner_audio.csv',',',1,0);

dims_set_genre1 = size(set_genre1);
n_genre1 = dims_set_genre1(1);

dims_set_genre2 = size(set_genre2);
n_genre2 = dims_set_genre2(1);

n_per_genre = min(n_genre1, n_genre2);

n_train_genre2 = floor(n_per_genre * 0.8);
n_crossval_genre2 = floor(n_per_genre * 0.1);
n_test_genre2 = floor(n_per_genre * 0.1);
n_train_genre1 = floor(n_per_genre * 0.8);
n_crossval_genre1 = floor(n_per_genre * 0.1);
n_test_genre1 = floor(n_per_genre * 0.1);



% Split sets into train, cross-validation and test data
% Remove first row, which contains the column names
data_train_genre1 = set_genre1(2:1+n_train_genre1, 3:end); 
data_test_genre1 = set_genre1(n_train_genre1 + 2 : n_train_genre1 + n_crossval_genre1 + n_test_genre1 + 1, 3:end); 

data_train_genre2 = set_genre2(1:n_train_genre2, 3:end); 
data_test_genre2 = set_genre2(n_train_genre2 + n_crossval_genre2 + 1 : n_train_genre2 + n_crossval_genre2 + n_test_genre2, 3:end); 


%% Entrainement du modele
n_train_per_genre = min(n_train_genre2, n_train_genre1);

% select certain features to plot 2 by 2 graph
X = [[data_train_genre1(1:n_train_per_genre, 3:4)]', [data_train_genre2(1:n_train_per_genre, 3:4)]']'; 
[X, mu, sigma] = featureNormalize(X);

y = [zeros(1, n_train_per_genre), ones(1, n_train_per_genre)]';
plotData(X,y);
xlim([-0.2 0.5]);

C = 1000; 
sigma = 0.3;
x1 = [-0.13 -0.135 -0.135 -0.14]; 
x2 = [0 1.5 -0.5 0]; 

model= svmTrain(X, y, C, @(x1, x2) gaussianKernel(x1, x2, sigma)); 
visualizeBoundary(X, y, model);