-
Notifications
You must be signed in to change notification settings - Fork 3
/
ATT.m
140 lines (113 loc) · 4.24 KB
/
ATT.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
function [PER,tt_core_es,Xre] = ATT(X_miss,Omega,tt_rank,OPTS_PER)
% Tensor Tracking with Missing Data Under Tensor-Train Format
% Author : LE Trung-Thanh
% Affiliation: University of Orleans, France
% Contact : trung-thanh.le@univ-orleans.fr // letrungthanhtbt@gmail.com
% Date : 09/01/2022
%%
tt_dim = size(X_miss);
N = length(tt_dim);
T = tt_dim(N);
Xre = zeros(tt_dim);
PER = zeros(1,T);
if isfield(OPTS_PER,'window'), % Forgetting factor
window = OPTS_PER.window;
else window = 1;
end
if isfield(OPTS_PER,'forget_factor'), % Forgetting factor
beta = OPTS_PER.forget_factor;
else beta = 0.5;
end
if isfield(OPTS_PER,'rho'), % Forgetting factor
rho = OPTS_PER.rho;
else rho = 1;
end
Xtrue = OPTS_PER.Xtrue;
%% Initialization
G{1} = randn(tt_dim(1),tt_rank(1));
G{2} = randn(tt_rank(1),tt_dim(2),tt_rank(2));
G{3} = randn(tt_rank(2),tt_dim(3),tt_rank(3));
G{4} = [];
Delta_G{1} = zeros(tt_dim(1),tt_rank(1));
Delta_G{2} = zeros(tt_dim(2),tt_rank(1)*tt_rank(2));
Delta_G{3} = zeros(tt_dim(3),tt_rank(2)*tt_rank(3));
tt_core_es = cell(N,1);
S_re{1} = 1*eye(tt_rank(1));
S_re{2} = 1*eye(tt_rank(1)*tt_rank(2));
S_re{3} = 1*eye(tt_rank(2)*tt_rank(3));
ii = 1;
while ii <= T
%% Data Collection
if ii + window - 1 > T
t = ii : T;
window = T - ii + 1;
else
t = ii : ii + window - 1;
end
X_t = X_miss(:,:,:,t);
Omega_t = Omega(:,:,:,t);
%% The last G4
G4_buffer = tt_product_tensors(tt_product_tensors(G{1},G{2}),G{3});
H_t = ten2mat(tensor(G4_buffer),4)';
G4 = [];
Delta_X_t = tensor(zeros([tt_dim(1:end-1) window]));
for ll = 1 : window
if window == 1
X_t_ll = X_t;
else
X_t_ll = X_t(:,:,:,ll);
end
x_t_ll = X_t_ll.data(:);
Omega_t_ll = Omega_t(:,:,:,ll);
omega_t_ll = Omega_t_ll(:);
idx_t = find(omega_t_ll);
H_Omega = H_t(idx_t,:);
g4 = H_Omega \ x_t_ll(idx_t);
G4 = [G4 g4];
X_t_ll_re = H_t * g4;
X_t_ll_re = tensor(reshape(X_t_ll_re,[tt_dim(1:end-1)]));
Delta_X_t_ll = Omega_t_ll .* (X_t_ll - X_t_ll_re);
Delta_X_t(:,:,:,ll) = Delta_X_t_ll;
end
G{4} = [G{4}, G4];
%% G1
ER_Unfolding_1 = ten2mat(Delta_X_t,1);
G_buffer{1} = ttm(tensor(tt_product_tensors(G{2},G{3})),G4',4);
W{1} = ten2mat(G_buffer{1},1);
S_re{1} = beta*S_re{1} + W{1}*W{1}';
Delta_G{1} = (ER_Unfolding_1 * W{1}' + beta*rho*Delta_G{1}) ...
* ( inv(S_re{1} + rho* eye(tt_rank(1))) )';
G{1} = G{1} + Delta_G{1};
%% G2
ER_Unfolding_2 = ten2mat(Delta_X_t,2);
G_buffer{2} = ttm(tensor(G{3}),G4',3);
G_buffer{2} = ten2mat(G_buffer{2},1);
W{2} = kron(G_buffer{2}, G{1}');
S_re{2} = beta*S_re{2} + W{2}*W{2}';
Delta_G{2} = (ER_Unfolding_2 * W{2}' + beta*rho*Delta_G{2}) ...
* ( inv(S_re{2} + rho*eye(tt_rank(1)*tt_rank(2))) )';
G2_Unfolding_2 = ten2mat(tensor(G{2}),2) + Delta_G{2};
G{2} = mat2ten(G2_Unfolding_2,[tt_rank(1), tt_dim(2), tt_rank(2)],[2]);
%% G3
ER_Unfolding_3 = ten2mat(Delta_X_t,3);
G_buffer{3} = ten2mat(tensor(tt_product_tensors(G{1},G{2})),3);
W{3} = kron(G4,G_buffer{3});
S_re{3} = beta*S_re{3} + W{3} * W{3}';
G3_Unfolding_2 = ten2mat(tensor(G{3}),2);
Delta_G{3} = (ER_Unfolding_3 * W{3}' + beta*rho*Delta_G{3}) ...
* ( inv(S_re{3} + rho*eye(tt_rank(2)*tt_rank(3))) )';
G3_Unfolding_2 = G3_Unfolding_2 + Delta_G{3};
G{3} = mat2ten(G3_Unfolding_2,[tt_rank(2), tt_dim(3), tt_rank(3)],[2]);
%% Save
tt_core_es{1} = G{1};
tt_core_es{2} = G{2};
tt_core_es{3} = G{3};
tt_core_es{4} = G4';
%% Performance estimation
X_t_true = Xtrue(:,:,:,ii);
X_t_re = tt_recover_tensor(tt_core_es);
X_t_re = X_t_re(:,:,:,1);
PER(1,t) = norm(X_t_true - X_t_re) / norm(X_t_true);
ii = ii + window;
end
end