IT++ Logo
mog_diag.cpp
Go to the documentation of this file.
1 
29 #include <itpp/base/math/log_exp.h>
30 #include <itpp/stat/mog_diag.h>
31 #include <cstdlib>
32 
33 
34 namespace itpp
35 {
36 
37 double MOG_diag::log_lhood_single_gaus_internal(const double * c_x_in, const int k) const
38 {
39 
40  const double * c_mean = c_means[k];
41  const double * c_diag_cov_inv_etc = c_diag_covs_inv_etc[k];
42 
43  double acc = 0.0;
44 
45  for (int d = 0; d < D; d++) {
46  double tmp_val = c_x_in[d] - c_mean[d];
47  acc += (tmp_val * tmp_val) * c_diag_cov_inv_etc[d];
48  }
49  return(c_log_det_etc[k] - acc);
50 }
51 
52 
53 double MOG_diag::log_lhood_single_gaus_internal(const vec &x_in, const int k) const
54 {
55  return log_lhood_single_gaus_internal(x_in._data(), k);
56 }
57 
58 
59 double MOG_diag::log_lhood_single_gaus(const double * c_x_in, const int k) const
60 {
61  if (do_checks) {
62  it_assert(valid, "MOG_diag::log_lhood_single_gaus(): model not valid");
63  it_assert(((k >= 0) && (k < K)), "MOG::log_lhood_single_gaus(): k specifies a non-existant Gaussian");
64  }
65  return log_lhood_single_gaus_internal(c_x_in, k);
66 }
67 
68 
69 double MOG_diag::log_lhood_single_gaus(const vec &x_in, const int k) const
70 {
71  if (do_checks) {
72  it_assert(valid, "MOG_diag::log_lhood_single_gaus(): model not valid");
73  it_assert(check_size(x_in), "MOG_diag::log_lhood_single_gaus(): x has wrong dimensionality");
74  it_assert(((k >= 0) && (k < K)), "MOG::log_lhood_single_gaus(): k specifies a non-existant Gaussian");
75  }
76  return log_lhood_single_gaus_internal(x_in._data(), k);
77 }
78 
79 
80 double MOG_diag::log_lhood_internal(const double * c_x_in)
81 {
82 
83  bool danger = paranoid;
84 
85  for (int k = 0;k < K;k++) {
86  double tmp = c_log_weights[k] + log_lhood_single_gaus_internal(c_x_in, k);
87  c_tmpvecK[k] = tmp;
88 
89  if (tmp >= log_max_K) danger = true;
90  }
91 
92 
93  if (danger) {
94  double log_sum = c_tmpvecK[0];
95  for (int k = 1; k < K; k++) log_sum = log_add(log_sum, c_tmpvecK[k]);
96  return(log_sum);
97  }
98  else {
99  double sum = 0.0;
100  for (int k = 0;k < K;k++) sum += std::exp(c_tmpvecK[k]);
101  return(std::log(sum));
102  }
103 }
104 
105 
106 double MOG_diag::log_lhood_internal(const vec &x_in)
107 {
108  return log_lhood_internal(x_in._data());
109 }
110 
111 
112 double MOG_diag::log_lhood(const vec &x_in)
113 {
114  if (do_checks) {
115  it_assert(valid, "MOG_diag::log_lhood(): model not valid");
116  it_assert(check_size(x_in), "MOG_diag::log_lhood(): x has wrong dimensionality");
117  }
118  return log_lhood_internal(x_in._data());
119 }
120 
121 
122 double MOG_diag::log_lhood(const double * c_x_in)
123 {
124  if (do_checks) {
125  it_assert(valid, "MOG_diag::log_lhood(): model not valid");
126  it_assert((c_x_in != 0), "MOG_diag::log_lhood(): c_x_in is a null pointer");
127  }
128 
129  return log_lhood_internal(c_x_in);
130 }
131 
132 
133 double MOG_diag::lhood_internal(const double * c_x_in)
134 {
135 
136  bool danger = paranoid;
137 
138  for (int k = 0;k < K;k++) {
139  double tmp = c_log_weights[k] + log_lhood_single_gaus_internal(c_x_in, k);
140  c_tmpvecK[k] = tmp;
141 
142  if (tmp >= log_max_K) danger = true;
143  }
144 
145 
146  if (danger) {
147  double log_sum = c_tmpvecK[0];
148  for (int k = 1; k < K; k++) log_sum = log_add(log_sum, c_tmpvecK[k]);
149  return(trunc_exp(log_sum));
150  }
151  else {
152  double sum = 0.0;
153  for (int k = 0;k < K;k++) sum += std::exp(c_tmpvecK[k]);
154  return(sum);
155  }
156 }
157 
158 double MOG_diag::lhood_internal(const vec &x_in) { return lhood_internal(x_in._data()); }
159 
160 double MOG_diag::lhood(const vec &x_in)
161 {
162  if (do_checks) {
163  it_assert(valid, "MOG_diag::lhood(): model not valid");
164  it_assert(check_size(x_in), "MOG_diag::lhood(): x has wrong dimensionality");
165  }
166  return lhood_internal(x_in._data());
167 }
168 
169 
170 double MOG_diag::lhood(const double * c_x_in)
171 {
172  if (do_checks) {
173  it_assert(valid, "MOG_diag::lhood(): model not valid");
174  it_assert((c_x_in != 0), "MOG_diag::lhood(): c_x_in is a null pointer");
175  }
176 
177  return lhood_internal(c_x_in);
178 }
179 
180 
181 double MOG_diag::avg_log_lhood(const double ** c_x_in, const int N)
182 {
183  if (do_checks) {
184  it_assert(valid, "MOG_diag::avg_log_lhood(): model not valid");
185  it_assert((c_x_in != 0), "MOG_diag::avg_log_lhood(): c_x_in is a null pointer");
186  it_assert((N >= 0), "MOG_diag::avg_log_lhood(): N is zero or negative");
187  }
188 
189  double acc = 0.0;
190  for (int n = 0;n < N;n++) acc += log_lhood_internal(c_x_in[n]);
191  return(acc / N);
192 }
193 
194 
196 {
197  if (do_checks) {
198  it_assert(valid, "MOG_diag::avg_log_lhood(): model not valid");
199  it_assert(check_size(X_in), "MOG_diag::avg_log_lhood(): X is empty or at least one vector has the wrong dimensionality");
200  }
201  const int N = X_in.size();
202  double acc = 0.0;
203  for (int n = 0;n < N;n++) acc += log_lhood_internal(X_in(n)._data());
204  return(acc / N);
205 }
206 
208 {
209  c_means = 0;
210  c_diag_covs = 0;
212  c_weights = 0;
213  c_log_weights = 0;
214  c_log_det_etc = 0;
215  c_tmpvecK = 0;
216 }
217 
218 
220 {
227  c_tmpvecK = disable_c_access(c_tmpvecK);
228 }
229 
230 
232 {
236 }
237 
238 
240 {
242  if (full) return;
243 
247 
251 }
252 
253 
255 {
257 
260 
263 }
264 
265 
267 {
268  disable_c_access(c_tmpvecK);
269  tmpvecK.set_size(K);
270  c_tmpvecK = enable_c_access(tmpvecK);
271 
274 }
275 
276 
277 void MOG_diag::load(const std::string &name_in)
278 {
279  MOG_generic::load(name_in);
280  if (full) convert_to_diag();
281 }
282 
283 
285 {
286  int rows = A_in.size();
287  double ** A = (double **)std::malloc(rows * sizeof(double *));
288  if (A) for (int row = 0;row < rows;row++) A[row] = A_in(row)._data();
289  return(A);
290 }
291 
293 {
294  int rows = A_in.size();
295  int ** A = (int **)std::malloc(rows * sizeof(int *));
296  if (A) for (int row = 0;row < rows;row++) A[row] = A_in(row)._data();
297  return(A);
298 }
299 
300 double ** MOG_diag::disable_c_access(double ** A_in) { if (A_in) std::free(A_in); return(0); }
301 int ** MOG_diag::disable_c_access(int ** A_in) { if (A_in) std::free(A_in); return(0); }
302 
303 double * MOG_diag::enable_c_access(vec & v_in) { return v_in._data(); }
304 int * MOG_diag::enable_c_access(ivec & v_in) { return v_in._data(); }
305 
306 double * MOG_diag::disable_c_access(double *) { return(0); }
307 int * MOG_diag::disable_c_access(int *) { return(0); }
308 
309 }
double log_lhood_single_gaus(const double *c_x_in, const int k) const
calculate the log likelihood of C vector c_x_in using only Gaussian k
Definition: mog_diag.cpp:59
Diagonal Mixture of Gaussians class - header file.
double log_add(double log_a, double log_b)
Safe substitute for log(exp(log_a) + exp(log_b))
Definition: log_exp.cpp:40
void convert_to_diag_internal()
ADD DOCUMENTATION HERE.
int size() const
Returns the number of data elements in the array object.
Definition: array.h:155
int D
dimensionality
Definition: mog_generic.h:295
virtual void setup_covs()
additional processing of covariance vectors/matrices, done as the last step of covariance initialisat...
Array< vec > means
means
Definition: mog_generic.h:298
void load(const std::string &name_in)
Initialise the model by loading the parameters from a model file.
Definition: mog_diag.cpp:277
double * c_log_weights
pointer to the log version of the weight vector
Definition: mog_diag.h:209
double * c_log_det_etc
pointer to the log_det_etc vector
Definition: mog_diag.h:212
void setup_misc()
additional processing of miscellaneous parameters, done as the last step of overall initialisation ...
Definition: mog_diag.cpp:266
double ** c_diag_covs
pointers to the covariance vectors
Definition: mog_diag.h:200
T sum(const Vec< T > &v)
Sum of all elements in the vector.
Definition: matfunc.h:59
double log_lhood_internal(const double *c_x_in)
ADD DOCUMENTATION HERE.
Definition: mog_diag.cpp:80
#define it_assert(t, s)
Abort if t is not true.
Definition: itassert.h:94
Logarithmic and exponenential functions - header file.
virtual void setup_misc()
additional processing of miscellaneous parameters, done as the last step of overall initialisation ...
double ** c_means
pointers to the mean vectors
Definition: mog_diag.h:197
bool check_size(const vec &x_in) const
Check if vector x_in has the same dimensionality as the model.
vec log(const vec &x)
The natural logarithm of the elements.
Definition: log_exp.h:241
virtual void setup_weights()
additional processing of the weight vector, done as the last step of weight initialisation ...
double lhood_internal(const double *c_x_in)
ADD DOCUMENTATION HERE.
Definition: mog_diag.cpp:133
double lhood(const double *c_x_in)
calculate the likelihood of C vector c_x_in
Definition: mog_diag.cpp:170
vec log_weights
Pre-calculated log versions of the weights.
Definition: mog_generic.h:320
vec exp(const vec &x)
Exp of the elements of a vector x.
Definition: log_exp.h:155
bool full
indicates whether we are using full or diagonal covariance matrices
Definition: mog_generic.h:286
Array< vec > diag_covs_inv_etc
Pre-calcuated inverted version of each diagonal covariance vector, where the covariance elements are ...
Definition: mog_generic.h:326
void free_all_ptrs()
ADD DOCUMENTATION HERE.
Definition: mog_diag.cpp:219
bool do_checks
indicates whether checks on input data are done
Definition: mog_generic.h:280
double ** disable_c_access(double **A_in)
Disable C style access to an Array of vectors (vec)
Definition: mog_diag.cpp:300
double log_lhood_single_gaus_internal(const double *c_x_in, const int k) const
ADD DOCUMENTATION HERE.
Definition: mog_diag.cpp:37
virtual void convert_to_diag()
Convert the model to use diagonal covariances.
itpp namespace
Definition: itmex.h:36
double ** c_diag_covs_inv_etc
pointers to the inverted covariance vectors
Definition: mog_diag.h:203
virtual void setup_means()
additional processing of mean vectors, done as the last step of mean initialisation ...
double ** enable_c_access(Array< vec > &A_in)
Enable C style access to an Array of vectors (vec)
Definition: mog_diag.cpp:284
void setup_covs()
additional processing of covariance vectors/matrices, done as the last step of covariance initialisat...
Definition: mog_diag.cpp:239
void setup_means()
additional processing of mean vectors, done as the last step of mean initialisation ...
Definition: mog_diag.cpp:231
virtual void load(const std::string &name_in)
Initialise the model by loading the parameters from a model file.
vec log_det_etc
Gaussian specific pre-calcualted constants.
Definition: mog_generic.h:317
double avg_log_lhood(const double **c_x_in, int N)
calculate the average log likelihood of an array of C vectors ( c_x_in )
Definition: mog_diag.cpp:181
double trunc_exp(double x)
Truncated exponential function.
Definition: log_exp.h:137
bool paranoid
indicates whether we are paranoid about numerical stability
Definition: mog_generic.h:289
void setup_weights()
additional processing of the weight vector, done as the last step of weight initialisation ...
Definition: mog_diag.cpp:254
int K
number of gaussians
Definition: mog_generic.h:292
bool valid
indicates whether the parameters are valid
Definition: mog_generic.h:283
Array< vec > diag_covs
diagonal covariance matrices, stored as vectors
Definition: mog_generic.h:301
double log_max_K
Pre-calcualted std::log(std::numeric_limits<double>::max() / K), where K is the number of Gaussians...
Definition: mog_generic.h:310
double * c_weights
pointer to the weight vector
Definition: mog_diag.h:206
vec weights
weights
Definition: mog_generic.h:307
double log_lhood(const double *c_x_in)
calculate the log likelihood of C vector c_x_in
Definition: mog_diag.cpp:122
void zero_all_ptrs()
ADD DOCUMENTATION HERE.
Definition: mog_diag.cpp:207
SourceForge Logo

Generated on Sun Apr 10 2022 12:00:00 for IT++ by Doxygen 1.8.14