-
Notifications
You must be signed in to change notification settings - Fork 1
/
distribution_anomaly.m
281 lines (252 loc) · 9.54 KB
/
distribution_anomaly.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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
RUN_GRUBBS_TEST = 0
RUN_Mahalanobis = 0
RUN_KLDivergence = 0
RUN_EMD = 1
RUN_MEAN = 0
USE_GMM = 1
CASE=1
if CASE == 1
% filepath = '/home/chenchu/volumes/TEMP/14.20_16.00_14.20x2_150523/'; id_start = 49401
% filepath = '/home/chenchu/volumes/TEMP/13.80_141219/'; id_start = 6201
filepath = '/home/chenchu/volumes/TEMP/14.20_16.00/'; id_start = 20601
filepattern = 's35_noinj.r2b%d.p3d.g%d_Pressure.raw'
output_filepattern_grubbs = 'distribution_diff/distdiff_b%d_%d.raw'
output_filepattern_mahalanobis = 'distribution_diff/mahalanobis_b%d_%d.raw'
output_filepattern_KLD = 'distribution_diff/kldiv_b%d_%d.raw'
output_filepattern_EMD = 'distribution_diff/emd_b%d_%d.raw'
output_filepattern_mean = 'distribution_diff/mean_b%d_%d.raw'
id_step = 25
nfiles = 576
W=151; % Note!! Original range
H=71
D=56;
res = 5;
end
NBINS = 64;
nblocks = 36;
maxval = 4;
edges = linspace(0, maxval, NBINS+1);
w1 = floor(W/res);
h1 = floor(H/res);
d1 = floor(D/res);
ncells = w1*h1*d1
Z_TH = 3.3296;
% center of each bin:
bin_val = (edges(1:end-1)+edges(2:end)) * 0.5; bin_val = bin_val';
% preallocate necessary buffers
histmat = zeros(NBINS, ncells, nblocks);
anomaly = zeros(ncells,nblocks);
anomaly_count = zeros(ncells,nblocks);
gmmmat = cell(ncells, nblocks);
for t=1:nfiles
for b=1:nblocks
filename = sprintf(strcat(filepath, filepattern), b, id_start+id_step*(t-1) )
fp = fopen(filename, 'rb');
data = fread(fp, W*H*D, 'float32');
fclose(fp);
data = reshape(data, [W,H,D]);
%% compute histogram
count = 1;
for z=1:res:d1*res
for y=1:res:h1*res
for x=1:res:w1*res
tmp = data(x:x+res-1, ...
y:y+res-1, ...
z:z+res-1 );
tmp = reshape(tmp, [res*res*res,1]);
if USE_GMM
gmm = isp_gmmtrain(tmp', 'nMixtures', 4, 'silent', true);
gmmmat{count,b} = gmm;
%
% gmmmat{count,b} = gmdistribution.fit(tmp, 4, 'CovType','diagonal');
% % verfication:
% h = histc(tmp, edges);
% hold off
% plot( h )
% hold on
% plot( pdf(gmmmat{count,b}, edges') );
% cov = zeros(1,1,4);
% cov(1,1,:) = gmm.covariance;
% plot( pdf(gmdistribution(gmm.mean', cov, gmm.weight'), edges'))
% pause
% hold off
h = histc(tmp, edges);
histmat(:,count,b) = h(1:NBINS);
else
% hh = histcounts(tmp, edges);
% histmat(:,count,b) = hh';
h = histc(tmp, edges);
histmat(:,count,b) = h(1:NBINS);
end
count = count +1;
end
end
end
end
%% anomaly analysis
if RUN_GRUBBS_TEST
for i=1:ncells
cell_hist = sum( histmat(:,i,:), 3 );
npoints = sum(cell_hist);
cell_mean = sum(bin_val.*cell_hist)/npoints;
cell_std = sqrt(sum(bin_val.*bin_val.*cell_hist)/npoints - cell_mean*cell_mean); % sqrt(E(X^2)-E(X)^2)
anomaly_bins = (bin_val > cell_mean + cell_std * Z_TH + maxval/NBINS*.5) | ...
(bin_val < cell_mean - cell_std * Z_TH - maxval/NBINS*.5);
for b=1:nblocks
anomaly_count(i, b) = sum(anomaly_bins .* histmat(:,i,b) );
% if (anomaly_count(i,b) > 0)
% anomaly_count(i,b)
% % pause
% end
end
end
for b=1:nblocks
filename = strcat(filepath, sprintf(output_filepattern_grubbs, b, t) )
fp = fopen(filename, 'wb')
fwrite(fp, anomaly_count(:,b), 'float32');
fclose(fp);
end
end
% regard each distribution as a high-dim vector
if RUN_Mahalanobis
for i=1:ncells
mat = squeeze(histmat(:,i,:)); % [bins blocks]
c = mean(mat, 2);
matc = mat - repmat(c, 1, nblocks);
% nonzero rows
rows = find(all(matc,2));
% all 0?
if (length(rows) == 0)
anomaly(i,:) = zeros(1, nblocks);
continue;
end
matc1 = matc(rows, :);
cov = matc1*matc1'/nblocks;
% svd
if 0
[U,S,V] = svd(cov);
s=diag(S);
s(find(s==0)) = 1;
for b=1:nblocks
offset = matc1(:,b);
anomaly(i,b) = sqrt( offset'*V*(U'*offset./s) );
end
end
% eigen analysis. faster than svd
if 1
[PC, V] = eig(cov);
V=diag(V);
V(find(V==0)) = 1;
for b=1:nblocks
offset = matc1(:,b);
anomaly(i,b) = sqrt( sum((PC' * offset).^2 ./ V ) );
end
end
% original definition. not stable
if 0
invcov = inv(cov);
a= zeros(1,b);
for b=1:nblocks
offset = matc1(:,b);
a(b) = sqrt(offset'*invcov*offset);
end
a
end
end
for b=1:nblocks
filename = strcat(filepath, sprintf(output_filepattern_mahalanobis, b, t) )
fp = fopen(filename, 'wb')
fwrite(fp, anomaly(:,b), 'float32');
fclose(fp);
end
end
% regard each distribution as a high-dim vector, and compare with
if RUN_KLDivergence
for i=1:ncells
mat = squeeze(histmat(:,i,:)); % [bins blocks]
mat = mat/(res*res*res); % get probability
% mean distribution
q = mean(mat, 2);
q(q==0) = 1; % esacpe 0
qmat = repmat(q, 1, nblocks);
mat1 = mat; mat1(mat1==0)=1; % escape 0
d = sum(mat .* log(mat1 ./ qmat), 1);
% imagesc([mat q; d 1])
anomaly(i,:) = d;
end
for b=1:nblocks
filename = strcat(filepath, sprintf(output_filepattern_KLD, b, t) )
fp = fopen(filename, 'wb');
fwrite(fp, anomaly(:,b), 'float32');
fclose(fp);
end
end
if RUN_EMD
for i=1:ncells
if ~USE_GMM
mat = squeeze(histmat(:,i,:)); % [bins blocks]
mat = mat/(res*res*res); % get probability
cummat = cumsum(mat, 1); % CDF
% mean distribution
c = mean(mat, 2);
cumc = cumsum(c);
cumcmat = repmat(cumc, 1, nblocks);
anomaly(i, :) = sum( abs(cumcmat-cummat) ); % L1
% anomaly(i, :) = sqrt(sum( (cumcmat-cummat).^2 )); % L2
% imagesc([mat c; anomaly(i,:) 0]);
% pause
else
means = zeros(1,36*4);
vars = zeros(1,36*4);
weights = zeros(1,36*4);
for b=1:36
gmm = gmmmat{i,b}
means((b-1)*4+1 : (b-1)*4+4) = gmm.mean;
vars((b-1)*4+1 : (b-1)*4+4) = gmm.covariance;
weights((b-1)*4+1 : (b-1)*4+4) = gmm.weight;
end
gmm.mean = means;
gmm.covariance = vars;
gmm.weight = weights/36;
gmm.nMixtures = 4*36;
for b=1:36
d = isp_gmmdistance(gmm, gmmmat{i,b})
anomaly(i,b)=d;
end
% verify
mat = squeeze(histmat(:,i,:)); % [bins blocks]
mat = mat/(res*res*res); % get probability
c = mean(mat, 2);
cummat = cumsum(mat, 1); % CDF
imagesc([mat c; anomaly(i,:) 0]);
pause
end
end
for b=1:nblocks
filename = strcat(filepath, sprintf(output_filepattern_EMD, b, t) )
fp = fopen(filename, 'wb')
fwrite(fp, anomaly(:,b), 'float32');
fclose(fp);
end
end
if RUN_MEAN
pts = res*res*res;
for i=1:ncells
mat = squeeze(histmat(:,i,:)); % [bins blocks]
dist_average = sum(mat .* repmat(bin_val, 1, nblocks), 1) / pts;
s = std(dist_average);
m = mean(dist_average);
if s==0
anomaly(i,:) = zeros(1, nblocks);
continue;
end
anomaly(i,:) = (dist_average' - m)/s;
end
for b=1:nblocks
filename = strcat(filepath, sprintf(output_filepattern_mean, b, t) )
fp = fopen(filename, 'wb')
fwrite(fp, anomaly(:,b), 'float32');
fclose(fp);
end
end
end