wolffd@0
|
1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
2 % Experiment code templat %
|
wolffd@0
|
3 % Project: sub-euclidean distance for music similarity %
|
wolffd@0
|
4 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
5 %% Load features
|
wolffd@0
|
6 feature_file = 'rel_music_raw_features.mat';
|
wolffd@0
|
7 vars = whos('-file', feature_file);
|
wolffd@0
|
8 A = load(feature_file,vars(1).name,vars(2).name,vars(3).name,vars(4).name);
|
wolffd@0
|
9 raw_features = A.(vars(1).name);
|
wolffd@0
|
10 indices = A.(vars(2).name);
|
wolffd@0
|
11 tst_inx = A.(vars(3).name);
|
wolffd@0
|
12 trn_inx = A.(vars(4).name);
|
wolffd@0
|
13 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
14 % Define directory to save parameters & results
|
wolffd@0
|
15 % dir = '/home/funzi/Documents/';
|
wolffd@0
|
16 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
17 dmr = [0 5 10 20 30 50]; % dimension reduction by PCA
|
wolffd@0
|
18 ws = [0 5 10 20 30 50 70]; % window size
|
wolffd@0
|
19 % parameters of rbm (if it is used for extraction)
|
wolffd@0
|
20 hidNum = [30 50 100 500];
|
wolffd@0
|
21 lr_1 = [0.05 0.1 0.5];
|
wolffd@0
|
22 lr_2 = [0.1 0.5 0.7];
|
wolffd@0
|
23 mmt = [0.02 0.05 0.1];
|
wolffd@0
|
24 cost = [0.00002 0.01 0.1];
|
wolffd@0
|
25
|
wolffd@0
|
26 %% Select parameters (if grid-search is not applied)
|
wolffd@0
|
27 di = 1;
|
wolffd@0
|
28 wi = 1;
|
wolffd@0
|
29 hi = 1;
|
wolffd@0
|
30 l1i = 1;
|
wolffd@0
|
31 l2i = 1;
|
wolffd@0
|
32 mi = 1;
|
wolffd@0
|
33 ci = 1;
|
wolffd@0
|
34 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
35 % If grid search is define
|
wolffd@0
|
36 % log_file = strcat(dir,'exp_.mat');
|
wolffd@0
|
37 % inx = resume_from_grid(log_file,8);
|
wolffd@0
|
38 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
39 %% Feature extraction
|
wolffd@0
|
40 EXT_TYPE = 2;
|
wolffd@0
|
41 switch (EXT_TYPE)
|
wolffd@0
|
42 case 1 % Using PCA
|
wolffd@0
|
43 assert(~exist('OCTAVE_VERSION'),'This script cannot run in octave');
|
wolffd@0
|
44 coeff = princomp(raw_features);
|
wolffd@0
|
45 coeff = coeff(:,1:end-dmr(di)); % Change value of dmr(di) to reduce the dimensionality
|
wolffd@0
|
46 features = raw_features*coeff;
|
wolffd@0
|
47 % normalizing
|
wolffd@0
|
48 mm = minmax(features')';
|
wolffd@0
|
49 inn= (find(mm(1,:)~=mm(2,:)));
|
wolffd@0
|
50 mm = mm(:,inn);
|
wolffd@0
|
51 features = features(:,inn);
|
wolffd@0
|
52 features = (features-repmat(mm(1,:),size(features,1),1))./(repmat(mm(2,:),size(features,1),1)-repmat(mm(1,:),size(features,1),1));
|
wolffd@0
|
53 case 2 % Using rbm
|
wolffd@0
|
54 conf.hidNum = hidNum(hi);
|
wolffd@0
|
55 conf.eNum = 100;
|
wolffd@0
|
56 conf.sNum = size(raw_features,1);
|
wolffd@0
|
57 conf.bNum = 1;
|
wolffd@0
|
58 conf.gNum = 1;
|
wolffd@0
|
59 conf.params = [lr_1(l1i) lr_2(l2i) mmt(mi) cost(ci)];
|
wolffd@0
|
60 conf.N = 50;
|
wolffd@0
|
61 conf.MAX_INC = 10;
|
wolffd@0
|
62 W1 = zeros(0,0);
|
wolffd@0
|
63 [W1 vB1 hB1] = training_rbm_(conf,W1,raw_features);
|
wolffd@0
|
64 features = raw_features*W1 + repmat(hB1,conf.sNum,1);
|
wolffd@0
|
65 end
|
wolffd@0
|
66
|
wolffd@0
|
67 %% Sub-euclidean computation
|
wolffd@0
|
68 num_case = size(trn_inx,1);
|
wolffd@0
|
69 trnd_12 = cell(1,num_case);
|
wolffd@0
|
70 trnd_13 = cell(1,num_case);
|
wolffd@0
|
71 tstd_12 = cell(1,num_case);
|
wolffd@0
|
72 tstd_13 = cell(1,num_case);
|
wolffd@0
|
73
|
wolffd@0
|
74 w = ws(wi);
|
wolffd@0
|
75
|
wolffd@0
|
76 % w = subspace window size
|
wolffd@0
|
77 if w == 0 % trnd_12 = d(a,b) , trnd_13= d(a,c)
|
wolffd@0
|
78 for i = 1:num_case % over all cross-validation folds (num_case)
|
wolffd@0
|
79 [trnd_12{i} trnd_13{i}] = simple_dist(trn_inx{i},features,indices);
|
wolffd@0
|
80 [tstd_12{i} tstd_13{i}] = simple_dist(tst_inx{i},features,indices);
|
wolffd@0
|
81 end
|
wolffd@0
|
82 else
|
wolffd@0
|
83 for i = 1:num_case % for w > 1
|
wolffd@0
|
84 [trnd_12{i} trnd_13{i}] = conv_euclidean_dist(trn_inx{i},features,indices,w,1); %% normalize is better than no normalize
|
wolffd@0
|
85 [tstd_12{i} tstd_13{i}] = conv_euclidean_dist(tst_inx{i},features,indices,w,1);
|
wolffd@0
|
86 end
|
wolffd@0
|
87 end
|
wolffd@0
|
88 %% Data preparation
|
wolffd@0
|
89 trn_dat1 = cell(1,num_case);
|
wolffd@0
|
90 trn_dat2 = cell(1,num_case);
|
wolffd@0
|
91 tst_dat1 = cell(1,num_case);
|
wolffd@0
|
92 tst_dat2 = cell(1,num_case);
|
wolffd@0
|
93
|
wolffd@0
|
94 for i=1:num_case
|
wolffd@0
|
95 %=> Compute hypothesis
|
wolffd@0
|
96 trn_dat1{i} = trnd_13{i} - trnd_12{i};
|
wolffd@0
|
97 trn_dat2{i} = trnd_12{i} - trnd_13{i};
|
wolffd@0
|
98 tst_dat1{i} = tstd_13{i} - tstd_12{i};
|
wolffd@0
|
99 tst_dat2{i} = tstd_12{i} - tstd_13{i};
|
wolffd@0
|
100
|
wolffd@0
|
101 % => Normalize using logistic (lost the range)
|
wolffd@0
|
102 % trn_dat1{i} = logistic(trn_dat1{i});
|
wolffd@0
|
103 % trn_dat2{i} = logistic(trn_dat2{i});
|
wolffd@0
|
104 % tst_dat1{i} = logistic(tst_dat1{i});
|
wolffd@0
|
105 % tst_dat2{i} = logistic(tst_dat2{i});
|
wolffd@0
|
106
|
wolffd@0
|
107 %=> Normalize using min-max
|
wolffd@0
|
108 % mm = minmax(trn_dat1{i}')';
|
wolffd@0
|
109 % inn= find(mm(1,:)~=mm(2,:));
|
wolffd@0
|
110 % mm = mm(:,inn);
|
wolffd@0
|
111 % trn_dat1{i} =
|
wolffd@0
|
112 % (trn_dat1{i}(:,inn)-repmat(mm(1,:),size(trn_dat1{i},1),1))./repmat(mm(2,:)-mm(1,:),size(trn_dat1{i},1),1);
|
wolffd@0
|
113 % mm = minmax(tst_dat1{i}')';
|
wolffd@0
|
114 % inn= find(mm(1,:)~=mm(2,:));
|
wolffd@0
|
115 % mm = mm(:,inn);
|
wolffd@0
|
116 % tst_dat1{i} = (tst_dat1{i}(:,inn)-repmat(mm(1,:),size(tst_dat1{i},1),1))./repmat(mm(2,:)-mm(1,:),size(tst_dat1{i},1),1);
|
wolffd@0
|
117 %
|
wolffd@0
|
118 % mm = minmax(trn_dat2{i}')';
|
wolffd@0
|
119 % inn= find(mm(1,:)~=mm(2,:));
|
wolffd@0
|
120 % mm = mm(:,inn);
|
wolffd@0
|
121 % trn_dat2{i} = (trn_dat2{i}(:,inn)-repmat(mm(1,:),size(trn_dat2{i},1),1))./repmat(mm(2,:)-mm(1,:),size(trn_dat2{i},1),1);
|
wolffd@0
|
122
|
wolffd@0
|
123 % mm = minmax(tst_dat2{i}')';
|
wolffd@0
|
124 % inn= find(mm(1,:)~=mm(2,:));
|
wolffd@0
|
125 % mm = mm(:,inn);
|
wolffd@0
|
126 % tst_dat2{i} = (tst_dat2{i}(:,inn)-repmat(mm(1,:),size(tst_dat2{i},1),1))./repmat(mm(2,:)-mm(1,:),size(tst_dat2{i},1),1);
|
wolffd@0
|
127
|
wolffd@0
|
128 % => normalize from [-1 1] to [0 1]
|
wolffd@0
|
129 trn_dat1{i} = (trn_dat1{i}+1)/2;
|
wolffd@0
|
130 trn_dat2{i} = (trn_dat2{i}+1)/2;
|
wolffd@0
|
131 tst_dat1{i} = (tst_dat1{i}+1)/2;
|
wolffd@0
|
132 tst_dat2{i} = (tst_dat2{i}+1)/2;
|
wolffd@0
|
133 end
|
wolffd@0
|
134 correct = 0; % correct rate
|
wolffd@0
|
135 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
136 %% CODE HERE %%
|
wolffd@0
|
137 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
138
|
wolffd@0
|
139 fprintf('Correct = %f\n',correct);
|
wolffd@0
|
140 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
141 % Using the logging function to save paramters
|
wolffd@0
|
142 % and the result for plotting or in grid search
|
wolffd@0
|
143 % logging(log_file,[i1 i2 i3 i4 i5 correct]);
|
wolffd@0
|
144 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
wolffd@0
|
145
|
wolffd@0
|
146 clear; |