-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathtest_comparison_final.m
174 lines (154 loc) · 7.26 KB
/
test_comparison_final.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
clear all;
close all;
clc;
rng('shuffle')
% THIS METHOD WORKS ONLY FOR FIVE FOLD CROSS VALIDATION. SO TUBE_SUB_TRAIN
% SHOULD BE ONLY MULTIPLE OF 5 (AND NOT 16,64 - MAKE IT 15,60, etc).
% TUBE_SUB_TEST can be anything.
%% load the training data based on the input
addpath(genpath('\\engin-labs.m.storage.umich.edu\aniketde\windat.v2\Desktop\sem 4 transfer learning\JMLR_experiments\imdb experiments\imdb_lgl\liblinear-1.95'))
str1 = 'tube_sub';
string_version = '_v1';
str = strcat(str1,string_version);
load(str);
N = length(datasets);
numberOfTrainingUser = 20; %5,10,15,20,25,30,35
% rand_perm = randperm(N,N);
% for ii = 1:N
% tube_sub{ii} = tube_sub{rand_perm(ii)};
% end
% str1 = 'tube_sub_v10';
% str2 = '.mat';
% train_file = strcat(str1,str2);
% save(train_file,'tube_sub');
%% Initialization
datasets_training_num = numberOfTrainingUser;
numberOfExamplesPerTask_grid = 100%ceil(logspace(1.3,2,10));
for nn = 1:length(numberOfExamplesPerTask_grid)
numberOfExamplesPerTask = numberOfExamplesPerTask_grid(nn);
bw_kde1_est = 390.6940;
bw_kde2_est = 0.1758 ;
bw_kde3_est = 575.4399;
cost_est = 10;
lambda = 1.2500e-05;
bw_kde1_log = logspace(-2,4,20);
bw_kde2_log = 0.1758;
bw_kde3_log = 575.4399;
cost_log = logspace(-1,1,10);
L = 100;
Q = 100;
numb_iter = 5;
datasets_training_num_array = (1:datasets_training_num)';
tube_src_num = length(datasets_training_num_array);
fold_cv = 5; %because of 5 fold cross-validation
size_cv = datasets_training_num/fold_cv; %because of 5 fold cross-validation
datasets_training_num_per_task = numberOfExamplesPerTask*ones(datasets_training_num,4);
for ii = 1:datasets_training_num
permrand = randperm(length(datasets{ii}.testy),numberOfExamplesPerTask);%1:N;
datasets{ii}.x = datasets{ii}.testx(permrand,:);
datasets{ii}.y = datasets{ii}.testy(permrand);
end
%% Data collect
[X, Y, xse] = util_mrg_datasets(datasets, datasets_training_num_array, datasets_training_num_per_task);
[x, y, xe,X_test_cv,Y_test_cv,X_train_cv,Y_train_cv,P_X_test,P_X_train] = util_mrg_datasets_CV(datasets, datasets_training_num_array, datasets_training_num_per_task,fold_cv,size_cv,X,Y);
%util_mrg_datasets_CV
%% estimation of bandwidth for pooling
err = pooled_CV(bw_kde1_log,cost_log,numb_iter,X_test_cv,Y_test_cv,X_train_cv,Y_train_cv,fold_cv,Q);
[~, idx] = min(err(:));
[r1, r2] = ind2sub(size(err),idx);
err(r1,r2)
bw_kde1_est = bw_kde1_log(r1);
cost_est = cost_log(r2);
str1 = '_coordinate_ascent_cv_estimates_pooled_diffn_';
str2 = num2str(numberOfExamplesPerTask);
str3 = '.mat';
str4= num2str(numberOfTrainingUser);
str = strcat(str4,string_version,str1,str2,str3);
save(str,'err','bw_kde1_log','bw_kde2_log','bw_kde3_log','cost_log','N','numberOfTrainingUser','fold_cv')
bw_kde1_est_pooling = bw_kde1_est;
cost_est_pooling = cost_est;
%% estimation of bandwidth for ktl first loop
bw_kde1_log = bw_kde1_est;
bw_kde2_log = logspace(-2,4,20);
bw_kde3_log = logspace(-2,4,20);
cost_log = cost_est;
err = transfer_CV(bw_kde1_log,bw_kde2_log,bw_kde3_log,cost_log,size_cv,numb_iter,X_test_cv,Y_test_cv,X_train_cv,Y_train_cv,fold_cv,L,Q,P_X_train,P_X_test,datasets);
str1 = '_coordinate_ascent_cv_estimates_diff_patients_transfer_v1_diffn_';
str2 = num2str(numberOfExamplesPerTask);
str3 = '.mat';
str4= num2str(numberOfTrainingUser);
str = strcat(str4,string_version,str1,str2,str3);
save(str,'err','bw_kde1_log','bw_kde2_log','bw_kde3_log','cost_log','N','numberOfTrainingUser','fold_cv')
ani_err = reshape(err,[length(bw_kde2_log), length(bw_kde3_log)]);
[~, idx] = min(err(:));
[~, r1, r2] = ind2sub(size(err),idx);
err(1,r1,r2);
bw_kde2_est = bw_kde2_log(r1);
bw_kde3_est = bw_kde3_log(r2);
ani_err(r1,r2);
%% estimation of bandwidth for ktl first loop
bw_kde1_log = logspace(-2,4,20);
bw_kde2_log = bw_kde2_est;
bw_kde3_log = bw_kde3_est;
cost_log = logspace(-1,1,10);
err = transfer_CV(bw_kde1_log,bw_kde2_log,bw_kde3_log,cost_log,size_cv,numb_iter,X_test_cv,Y_test_cv,X_train_cv,Y_train_cv,fold_cv,L,Q,P_X_train,P_X_test,datasets);
str1 = '_coordinate_ascent_cv_estimates_diff_patients_transfer_v2_diffn_';
str2 = num2str(numberOfExamplesPerTask);
str3 = '.mat';
str4= num2str(numberOfTrainingUser);
str = strcat(str4,string_version, str1,str2,str3);
save(str,'err','bw_kde1_log','bw_kde2_log','bw_kde3_log','cost_log','N','numberOfTrainingUser','fold_cv')
err = reshape(err,[length(bw_kde1_log), length(cost_log)]);
[val_idx, idx] = min(err(:));
[r1, r2] = ind2sub(size(err),idx);
bw_kde1_est = bw_kde1_log(r1);
cost_est = cost_log(r2);
err(r1,r2);
%% estimation of bandwidth for ktl first loop
bw_kde1_log = bw_kde1_est;
bw_kde2_log = logspace(-2,4,20);
bw_kde3_log = logspace(-2,4,20);
cost_log = cost_est;
err = transfer_CV(bw_kde1_log,bw_kde2_log,bw_kde3_log,cost_log,size_cv,numb_iter,X_test_cv,Y_test_cv,X_train_cv,Y_train_cv,fold_cv,L,Q,P_X_train,P_X_test,datasets);
str1 = '_coordinate_ascent_cv_estimates_diff_patients_transfer_v3_diffn_';
str2 = num2str(numberOfExamplesPerTask);
str3 = '.mat';
str4= num2str(numberOfTrainingUser);
str = strcat(str4,string_version,str1,str2,str3);
save(str,'err','bw_kde1_log','bw_kde2_log','bw_kde3_log','cost_log','N','numberOfTrainingUser','fold_cv')
ani_err = reshape(err,[length(bw_kde2_log), length(bw_kde3_log)]);
[~, idx] = min(err(:));
[~, r1, r2] = ind2sub(size(err),idx);
err(1,r1,r2);
bw_kde2_est = bw_kde2_log(r1);
bw_kde3_est = bw_kde3_log(r2);
ani_err(r1,r2);
%% comparison
rand_perm_test = 36:42;
L_grid = L;
D = 100;
for ii = datasets_training_num+1:rand_perm_test(end)
datasets_training_num_per_task(ii,:) = length(datasets{ii}.testy)*ones(1,4);
end
for jj = datasets_training_num+1:rand_perm_test(end)
permrand = randperm(length(datasets{jj}.testy),numberOfExamplesPerTask);%1:N;
datasets{jj}.x = datasets{jj}.testx(permrand,:);
datasets{jj}.y = datasets{jj}.testy(permrand);
end
[res_avg_test, ~,res_avg_train,~] = pooled_transfer_comparison(datasets_training_num,bw_kde1_est,bw_kde2_est,bw_kde3_est,cost_est,rand_perm_test,L_grid,Q,D,numb_iter,datasets,datasets_training_num_per_task);
str1 = '_liblinear_ktl_diffn_';
str2 = '.mat';
str3= num2str(numberOfTrainingUser);
str = strcat(str3,string_version,str1,str2);
res_avg_test_loop(nn) = res_avg_test;
res_avg_train_loop(nn) = res_avg_train;
save(str,'res_avg_test_loop','res_avg_train_loop')
[~, res_avg_test_pooled,~,res_avg_train_pooled] = pooled_transfer_comparison(datasets_training_num,bw_kde1_est_pooling,bw_kde2_est,bw_kde3_est,cost_est_pooling,rand_perm_test,L_grid,Q,D,numb_iter,datasets,datasets_training_num_per_task);
str1 = '_liblinear_pooling_diffn_';
str2 = '.mat';
str3= num2str(numberOfTrainingUser);
str = strcat(str3,string_version,str1,str2);
res_avg_test_pooled_loop(nn) = res_avg_test_pooled;
res_avg_train_pooled_loop(nn) = res_avg_train_pooled;
save(str,'res_avg_test_pooled_loop','res_avg_train_pooled_loop')
end