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
|
///////////////////////////////////////////////////////////////////////////////
// //
// This file is part of ModelBlocks. Copyright 2009, ModelBlocks developers. //
// //
// ModelBlocks is free software: you can redistribute it and/or modify //
// it under the terms of the GNU General Public License as published by //
// the Free Software Foundation, either version 3 of the License, or //
// (at your option) any later version. //
// //
// ModelBlocks is distributed in the hope that it will be useful, //
// but WITHOUT ANY WARRANTY; without even the implied warranty of //
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the //
// GNU General Public License for more details. //
// //
// You should have received a copy of the GNU General Public License //
// along with ModelBlocks. If not, see <http://www.gnu.org/licenses/>. //
// //
// ModelBlocks developers designate this particular file as subject to //
// the "Moses" exception as provided by ModelBlocks developers in //
// the LICENSE file that accompanies this code. //
// //
///////////////////////////////////////////////////////////////////////////////
#ifndef _NL_GAUSS__
#define _NL_GAUSS__
#include <vector>
#include <string>
#include <math.h>
#include "nl-cpt.h"
#include "nl-prob.h"
#include "nl-list.h"
using namespace std;
static const PDFVal MEAN_THRESHOLD = 0.01; //0.0001; //0.001
static const PDFVal VARIANCE_THRESHOLD = 0.01; //0.0001; //0
////////////////////////////////////////////////////////////////////////////////
//
// Diagonal Multivariate Gaussian Model
//
////////////////////////////////////////////////////////////////////////////////
template <class Y>
class DiagGauss1DModel : public Generic1DModel<Y,PDFVal> {
private:
// Member variables...
string sId;
bool bModeled;
PDFVal prInvPowSqrt2PI;
SimpleHash<Id<int>,PDFVal> aMeans;
SimpleHash<Id<int>,PDFVal> aVariances;
PDFVal prInvRootNormVariances;
PDFVal prProduct;
SimpleHash<Id<int>,PDFVal> algprNegHalfInvVariances;
public:
// Constructor / destructor methods...
DiagGauss1DModel ( ) : bModeled(false) { }
DiagGauss1DModel ( const string& s ) : sId(s), bModeled(false) { }
// Specification methods...
void precomputeVarianceTerms ( ) ;
PDFVal& setMean ( int i ) { return aMeans.set(i); }
PDFVal& setVariance ( int i ) { return aVariances.set(i); }
PDFVal& setInvRootNormVar ( ) { return prInvRootNormVariances; }
PDFVal& setNegHalfInvVar ( int i ) { return algprNegHalfInvVariances.set(i); }
// Extraction methods...
PDFVal getMean ( int i ) const { return aMeans.get(i); }
PDFVal getVariance ( int i ) const { return aVariances.get(i); }
PDFVal getInvRootNormVar ( ) const { return prInvRootNormVariances; }
PDFVal getNegHalfInvVar ( int i ) const { return algprNegHalfInvVariances.get(i); }
int getNumFeats ( ) const { return Y::getSize(); }
PDFVal getProb ( const Y& ) const ;
// Input / output methods...
bool readFields ( char*[], int ) ;
void writeFields ( FILE*, const string& ) const ;
};
////////////////////////////////////////
template <class Y>
inline void DiagGauss1DModel<Y>::precomputeVarianceTerms ( ) {
// Inverse square root of norm of variances...
setInvRootNormVar() = 1.0;
for ( int i=0; i<getNumFeats(); i++ ) setInvRootNormVar() *= 1.0/sqrt(getVariance(i));
// Negative half of inverse of variances...
for ( int i=0; i<getNumFeats(); i++ ) setNegHalfInvVar(i) = -1.0/(2.0*getVariance(i));
// Derived from variance terms...
prInvPowSqrt2PI = 1.0/pow(sqrt(2.0*M_PI),getNumFeats());
prProduct = prInvPowSqrt2PI * getInvRootNormVar();
bModeled = true;
}
////////////////////////////////////////
template <class Y>
inline PDFVal DiagGauss1DModel<Y>::getProb ( const Y& y ) const {
// fprintf(stderr,"--------------------\n");
// y.write(stderr);
// fprintf(stderr,"\n----------\n");
// writeFields(stderr,"");
assert(bModeled);
PDFVal logprob = 0.0;
for ( int i=0; i<getNumFeats(); i++ )
logprob += getNegHalfInvVar(i) * pow(y.get(i)-getMean(i),2.0);
// for ( int i=0; i<getNumFeats(); i++ )
// fprintf(stderr,"%d %g\n", i, getNegHalfInvVar(i) * pow(y.get(i)-getMean(i),2.0));
// fprintf(stderr,"----------> %g\n",prProduct * exp(logprob));
return ( prProduct * exp(logprob) ) ;
}
////////////////////////////////////////
template <class Y>
bool DiagGauss1DModel<Y>::readFields ( char* as[], int numFields ) {
if ( 0==strcmp(as[1],"m") && numFields>2 ) {
char* psT;
for(int i=0;i<getNumFeats();i++)
setMean(i)=atof(strtok_r((0==i)?as[2]:NULL,"_",&psT));
}
else if ( 0==strcmp(as[1],"v") && numFields>2 ) {
char* psT;
for(int i=0;i<getNumFeats();i++)
setVariance(i)=atof(strtok_r((0==i)?as[2]:NULL,"_",&psT));
}
else return false;
return true;
}
////////////////////////////////////////
template <class Y>
void DiagGauss1DModel<Y>::writeFields ( FILE* pf, const string& sPref ) const {
fprintf(pf,"%s m = ",sPref.c_str());
for(int i=0; i<getNumFeats(); i++) fprintf(pf,"%s%f",(0==i)?"":"_",getMean(i));
fprintf ( pf, "\n" ) ;
fprintf(pf,"%s v = ",sPref.c_str());
for(int i=0; i<getNumFeats(); i++) fprintf(pf,"%s%f",(0==i)?"":"_",getVariance(i));
fprintf ( pf, "\n" ) ;
}
////////////////////////////////////////////////////////////////////////////////
/*
template <class Y,class X>
class DiagGauss2DModel : public Generic2DModel<Y,X,PDFVal> {
private:
// Member variables...
string sId;
SimpleHash<X,DiagGauss1DModel<Y> > mMY_giv_X;
public:
// Constructor / destructor methods...
DiagGauss2DModel ( const string& s ) : sId(s) { }
// Extraction methods...
Prob getProb ( const Y& y, const X& x ) const { return mMY_giv_X.get(x).getProb(y); }
// Input / output methods...
bool readFields ( char* as[], int numFields ) {
////if ( as[0]!=sId ) return false; // HAVE TO CHECK IN CALLIN FN NOW
if ( 0==strcmp(as[1],"m") && numFields>3 )
for ( int i=0; i<numFields-3; i++ ) mMY_giv_X.set(X(as[2])).setMean(i) = atof(as[i+4]) ;
else if ( 0==strcmp(as[1],"v") && numFields>3 )
for ( int i=0; i<numFields-3; i++ ) mMY_giv_X.set(X(as[2])).setVariance(i) = atof(as[i+4]) ;
else return false;
return true;
}
void writeFields ( FILE* pf, const string& sPref ) const {
X x;
for(bool b=x.setFirst(); b; b=x.setNext())
{ fprintf(pf,"%s m ",sPref.c_str()); x.write(pf); fprintf(pf," =");
for(int i=0; i<Y::getSize(); i++) fprintf(pf," %f",mMY_giv_X.getProb(x).getMean(i));
fprintf ( pf, "\n" ) ; }
for(bool b=x.setFirst(); b; b=x.setNext())
{ fprintf(pf,"%s v ",sPref.c_str()); x.write(pf); fprintf(pf," =");
for(int i=0; i<Y::getSize(); i++) fprintf(pf," %f",mMY_giv_X.getProb(x).getVariance(i));
fprintf ( pf, "\n" ) ; }
}
};
////////////////////////////////////////////////////////////////////////////////
template <class Y,class X1,class X2>
class DiagGauss3DModel : public Generic3DModel<Y,X1,X2,PDFVal> {
private:
// Member variables...
string sId;
SimpleHash<Joint2DRV<X1,X2>,DiagGauss1DModel<Y> > mMY_giv_X1_X2;
public:
// Constructor / destructor methods...
DiagGauss3DModel ( const string& s ) : sId(s) { }
// Extraction methods...
Prob getProb ( const Y& y, const X1& x1, const X2& x2 ) const { return mMY_giv_X1_X2.get(x1,x2).getProb(y); }
// Input / output methods...
bool readFields ( char* as[], int numFields ) {
if ( as[0]!=sId ) return false;
if ( 0==strcmp(as[1],"m") && numFields>4 )
for ( int i=0; i<numFields-4; i++ ) mMY_giv_X1_X2.set(Joint2DRV<X1,X2>(X1(as[2]),X2(as[2]))).setMean(i) = atof(as[i+4]) ;
else if ( 0==strcmp(as[1],"v") && numFields>4 )
for ( int i=0; i<numFields-4; i++ ) mMY_giv_X1_X2.set(Joint2DRV<X1,X2>(X1(as[2]),X2(as[2]))).setVariance(i) = atof(as[i+4]) ;
else return false;
return true;
}
void writeFields ( FILE* pf, string& sPref ) const {
X1 x1; X2 x2;
for(bool b1=x1.setFirst(); b1; b1=x1.setNext()) {
for(bool b2=x2.setFirst(); b2; b2=x2.setNext())
{ fprintf(pf,"%s m ",sPref.c_str()); x1.write(pf); fprintf(pf," "); x2.write(pf); fprintf(pf," =");
for(int i=0; i<Y::getSize(); i++) fprintf(pf," %f",mMY_giv_X1_X2.get(Joint2DRV<X1,X2>(x1,x2)).getMean(i));
fprintf(pf,"\n"); }
for(bool b2=x2.setFirst(); b2; b2=x2.setNext())
{ fprintf(pf,"%s v ",sPref.c_str()); x1.write(pf); fprintf(pf," "); x2.write(pf); fprintf(pf," =");
for(int i=0; i<Y::getSize(); i++) fprintf(pf," %f",mMY_giv_X1_X2.get(Joint2DRV<X1,X2>(x1,x2)).getVariance(i));
fprintf(pf,"\n"); }
}
}
};
*/
////////////////////////////////////////////////////////////////////////////////
//
// Trainable Diagonal Multivariate Gaussian Model
//
////////////////////////////////////////////////////////////////////////////////
template <class Y>
class TrainableDiagGauss1DModel : public DiagGauss1DModel<Y> {
public:
TrainableDiagGauss1DModel ( ) : DiagGauss1DModel<Y>() { }
TrainableDiagGauss1DModel ( const string& s ) : DiagGauss1DModel<Y>(s) { }
// input / output methods...
void setFields ( const List<pair<const Y*,Prob> >& ) ;
};
////////////////////////////////////////
template <class Y>
void TrainableDiagGauss1DModel<Y>::setFields ( const List<pair<const Y*,Prob> >& lyp ) {
// For each dimension...
for ( int i=0; i<DiagGauss1DModel<Y>::getNumFeats(); i++ ) {
// Calc means...
double curMean = DiagGauss1DModel<Y>::getMean(i);
DiagGauss1DModel<Y>::setMean(i) = 0.0;
// For each Y...
for ( const ListedObject<pair<const Y*,Prob> >* pyp=lyp.getFirst(); pyp; pyp=lyp.getNext(pyp) ) {
const Y& y = *pyp->first; // data value
const Prob& prEmpY = pyp->second; // empirical prob
//printf("cal mean i=%d x1=%s x2=%s aaaprYpsi.get(yd,x1,x2)=%f\n", i, x1.getString(), x2.getString(), (double)aaaprYpsi.getProb(yd,x1,x2));
DiagGauss1DModel<Y>::setMean(i) += prEmpY * y.get(i);
}
// // If any change exceeds thresh, continue...
// if ( bShouldStop && ( curMean - DiagGauss1DModel<Y>::getMean(i) > MEAN_THRESHOLD ||
// curMean - DiagGauss1DModel<Y>::getMean(i) < -MEAN_THRESHOLD ) ) bShouldStop = false;
//printf("cal mean i=%d getMean(i)=%f\n", i, DiagGauss1DModel<Y>::getMean(i));
// Calc variances...
double curVar = DiagGauss1DModel<Y>::getVariance(i);
DiagGauss1DModel<Y>::setVariance(i) = 0.0;
// For each Y...
for ( const ListedObject<pair<const Y*,Prob> >* pyp=lyp.getFirst(); pyp; pyp=lyp.getNext(pyp) ) {
const Y& y = *pyp->first; // data value
const Prob& prEmpY = pyp->second; // empirical prob
//printf("cal var i=%d yd=%s %f %f %f\n", i, yd.getString(), aaaprYpsi.get(yd,x1,x2), getMean(x1,x2,i), yd.get(i));
DiagGauss1DModel<Y>::setVariance(i) += prEmpY * pow(DiagGauss1DModel<Y>::getMean(i)-y.get(i),2) ;
}
// // If any change exceeds thresh, continue...
// if ( bShouldStop && ( curVar - DiagGauss1DModel<Y>::getVariance(i) > VARIANCE_THRESHOLD ||
// curVar - DiagGauss1DModel<Y>::getVariance(i) < -VARIANCE_THRESHOLD ) ) bShouldStop = false;
// Avoid div by zero...
if (DiagGauss1DModel<Y>::getVariance(i) < 1.0) DiagGauss1DModel<Y>::setVariance(i) = 1.0;
//printf("cal variance i=%d var=%f\n", i, DiagGauss1DModel<Y>::getVariance(i));
}
DiagGauss1DModel<Y>::precomputeVarianceTerms();
}
#endif /*_NL_GAUSS__*/
|
