-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtestimputation.m
319 lines (315 loc) · 10 KB
/
testimputation.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
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
% % Copyright 2014 C. P. de Campos ([email protected]). All rights reserved.
% % This work is licensed under a Creative Commons
% % Attribution-Noncommercial-Share Alike 3.0 United States License
% % http://creativecommons.org/licenses/by-nc-sa/3.0/us/
%
% Function to run multiple imputation methods over a data matrix.
function [ntests, mse, acc, names, impdata, bnets, topsorts, modeorexpected,single ] = testimputation(filename,R,C,cols,marperc,mcarperc,verb,bdeu,classe,tipos)
if nargin < 10
tipos=1:100;
end
if nargin < 9
classe=[];
end
if nargin < 7
verb = 1;
end
if nargin < 8, bdeu=1; end;
if numel(filename) > 1
M = filename;
clear filename;
else
M=csvread(filename,R,C);
end
if numel(cols)>0
M=M(:,cols);
end
s=size(M);
% train matrix has some elements discarded either by MCAR or MAR
train = 1+M-repmat(min(M),s(1),1);
% MAR missing data
if marperc > 0
for j=1:s(2)
for i=2:s(1)
if train(i-1,j)==train(i,j) && rand > (1-marperc)
train(i,j)=nan;
end
end
end
end
% MCAR missing data
if mcarperc > 0
train(rand(1,numel(train)) > (1-mcarperc)) = nan;
end
train = train';
% test matrix is complete
test = 1+M-repmat(min(M),s(1),1);
test=test';
% define test cases from the matrix
ntests = numel(find(isnan(train) & ~isnan(test)));
V=test(find(isnan(train) & ~isnan(test)));
acc = [];
mse = [];
impdata = {};
bnets = {};
names = {};
single= {};
nn = 1;
% train the BN
cel=matcell(train);
if any(tipos<4)
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,20,20,'E',1,classe);
end
% BN followed by picking mode
if any(tipos==1)
datT(topsort,:)=imputation(bnet,dat(topsort,:),'M');
R = cellmat(datT,'M');
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
impdata{nn} = R;
bnets{nn}=bnet;
topsorts{nn}=topsort;
names{nn} = 'BN_mode';
modeorexpected{nn} = 'M';
if verb, disp(names{nn}); end;
nn = nn + 1;
end
% BN followed by picking expected value
if any(tipos==2)
R = cellmat(dat,'E');
impdata{nn} = R;
modeorexpected{nn} = 'E';
names{nn} = 'BN_expectedvalue';
bnets{nn}=bnet;
topsorts{nn}=topsort;
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
if any(tipos==3)
% BN followed by picking round of expected value
R = cellmat(dat,'R');
impdata{nn} = R;
modeorexpected{nn} = 'R';
names{nn} = 'BN_roundedexpectedvalue';
bnets{nn}=bnet;
topsorts{nn}=topsort;
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
% pick intermediate value
% R = repmat((max(train)+min(train))/2,s(1),1);
% impdata{nn} = R;
%names{nn} = 'intermediatevalue';
% if verb, disp(names{nn}); end;
% nn = nn + 1;
% % compute MSE
% VV=R(find(isnan(train) & ~isnan(test)));
% mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
% acc = [ acc, sum(V==VV)/ntests ];
% % pick rounded intermediate value
% R = repmat(round((max(train)+min(train))/2),s(1),1);
% impdata{nn} = R;
%names{nn} = 'roundedintermediatevalue';
% if verb, disp(names{nn}); end;
% nn = nn + 1;
% % compute MSE
% VV=R(find(isnan(train) & ~isnan(test)));
% mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
% acc = [ acc, sum(V==VV)/ntests ];
% use single mean imputation
R = train;
RR = train;
s=size(R);
singval=zeros(1,s(1));
singmode=zeros(1,s(1));
for i=1:s(1)
v=0;
nv=zeros(max(R(i,:)),1);
for j=1:s(2)
if ~isnan(R(i,j))
v = v + R(i,j);
nv(R(i,j)) = nv(R(i,j)) + 1;
end
end
v = v/sum(nv);
singval(i)=v;
[m,mm] = max(nv);
singmode(i)=mm;
for j=1:s(2)
if isnan(R(i,j))
% mean
R(i,j) = v;
% mode
RR(i,j) = mm;
end
end
end
if any(tipos==4)
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,1,1,'EE',1,classe);
impdata{nn} = R;
bnets{nn}=bnet;
single{nn}=singval;
modeorexpected{nn} = 'E';
topsorts{nn}=topsort;
names{nn} = 'singlemean';
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
if any(tipos==5)
% now the rounded version
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,1,1,'EE',1,classe);
R = round(R);
impdata{nn} = R;
single{nn}=singmode;
topsorts{nn}=topsort;
bnets{nn}=bnet;
modeorexpected{nn} = 'R';
names{nn} = 'roundedsinglemean';
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
if any(tipos==6)
% now the mode
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,1,1,'EM',1,classe);
bnets{nn}=bnet;
single{nn}=singmode;
topsorts{nn}=topsort;
impdata{nn} = RR;
modeorexpected{nn} = 'M';
names{nn} = 'singlemode';
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=RR(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
if any(tipos==7)
% train the BN
cel=matcell(train);
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,20,20,'M',1,classe);
% BN followed by picking mode
R = cellmat(dat,'M');
impdata{nn} = R;
modeorexpected{nn} = 'M';
names{nn} = 'modeBN_mode';
bnets{nn}=bnet;
topsorts{nn}=topsort;
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
if any(tipos==8)
% train the BN with Romero & Salmeron idea IMPUTE_BN
cel=matcell(train);
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,20,30,'RS',1,classe);
% BN followed by picking mode
R = cellmat(dat,'M');
impdata{nn} = R;
modeorexpected{nn} = 'M';
names{nn} = 'romeros';
bnets{nn}=bnet;
topsorts{nn}=topsort;
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
if any(tipos==9)
% train the BN with DA algo (converges to posterior distro if structure was fixed...)
cel=matcell(train);
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,50,20,'DA',1,classe);
% BN followed by picking mode
R = cellmat(dat,'M');
impdata{nn} = R;
modeorexpected{nn} = 'M';
names{nn} = 'DA';
bnets{nn}=bnet;
topsorts{nn}=topsort;
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
if any(tipos==10)
% train the BN with full (complete graph) DA algo (converges to posterior distro...)
cel=matcell(train);
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,20,1,'FLDA',1,classe);
% BN followed by picking mode
R = cellmat(dat,'M');
impdata{nn} = R;
modeorexpected{nn} = 'M';
names{nn} = 'FullDA';
bnets{nn}=bnet;
topsorts{nn}=topsort;
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
if any(tipos==11)
% train the BN with full (complete graph) EM algo (converges to MAP...)
cel=matcell(train);
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,20,1,'FLE',1,classe);
% BN followed by picking mode
R = cellmat(dat,'E');
impdata{nn} = R;
modeorexpected{nn} = 'E';
names{nn} = 'FullEM';
bnets{nn}=bnet;
topsorts{nn}=topsort;
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
if any(tipos==12)
% train the BN with full (complete graph) EM algo followed
% by mode
cel=matcell(train);
[bnet,ml,dat,topsort]=structureEM(cel,bdeu,20,1,'FLM',1,classe);
% BN followed by picking mode
datT(topsort,:)=imputation(bnet,dat(topsort,:),'M');
R = cellmat(datT,'M');
impdata{nn} = R;
modeorexpected{nn} = 'M';
names{nn} = 'FullEMMode';
bnets{nn}=bnet;
topsorts{nn}=topsort;
if verb, disp(names{nn}); end;
nn = nn + 1;
% compute MSE
VV=R(find(isnan(train) & ~isnan(test)));
mse = [ mse, sqrt(sum((V-VV)'*(V-VV))/ntests) ];
acc = [ acc, sum(V==VV)/ntests ];
end
end