SHOGUN  4.1.0
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
Kernel.h
Go to the documentation of this file.
1 /*
2  * EXCEPT FOR THE KERNEL CACHING FUNCTIONS WHICH ARE (W) THORSTEN JOACHIMS
3  * COPYRIGHT (C) 1999 UNIVERSITAET DORTMUND - ALL RIGHTS RESERVED
4  *
5  * this program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; either version 3 of the License, or
8  * (at your option) any later version.
9  *
10  * Written (W) 1999-2009 Soeren Sonnenburg
11  * Written (W) 1999-2008 Gunnar Raetsch
12  * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society
13  */
14 
15 #ifndef _KERNEL_H___
16 #define _KERNEL_H___
17 
18 #include <shogun/lib/config.h>
19 
20 #include <shogun/lib/common.h>
21 #include <shogun/lib/Signal.h>
22 #include <shogun/io/SGIO.h>
23 #include <shogun/io/File.h>
26 #include <shogun/base/SGObject.h>
27 #include <shogun/lib/SGMatrix.h>
30 
31 namespace shogun
32 {
33  class CFile;
34  class CFeatures;
36 
37 #ifdef USE_SHORTREAL_KERNELCACHE
38 
40 #else
41 
43 #endif
44 
46 typedef int64_t KERNELCACHE_IDX;
47 
48 
51 {
54 };
55 
58 {
59  K_UNKNOWN = 0,
60  K_LINEAR = 10,
61  K_POLY = 20,
62  K_GAUSSIAN = 30,
66  K_SALZBERG = 41,
74  K_POLYMATCH = 100,
75  K_ALIGNMENT = 110,
80  K_COMBINED = 140,
81  K_AUC = 150,
82  K_CUSTOM = 160,
83  K_SIGMOID = 170,
84  K_CHI2 = 180,
85  K_DIAG = 190,
86  K_CONST = 200,
87  K_DISTANCE = 220,
90  K_OLIGO = 250,
91  K_MATCHWORD = 260,
92  K_TPPK = 270,
96  K_WAVELET = 310,
97  K_WAVE = 320,
98  K_CAUCHY = 330,
99  K_TSTUDENT = 340,
103  K_SPHERICAL = 380,
104  K_SPLINE = 390,
105  K_ANOVA = 400,
106  K_POWER = 410,
107  K_LOG = 420,
108  K_CIRCULAR = 430,
111  K_BESSEL = 460,
113  K_DIRECTOR = 480,
114  K_PRODUCT = 490,
115  K_LINEARARD = 500,
119 };
120 
123 {
124  KP_NONE = 0,
125  KP_LINADD = 1, // Kernels that can be optimized via doing normal updates w + dw
126  KP_KERNCOMBINATION = 2, // Kernels that are infact a linear combination of subkernels K=\sum_i b_i*K_i
127  KP_BATCHEVALUATION = 4 // Kernels that can on the fly generate normals in linadd and more quickly/memory efficient process batches instead of single examples
128 };
129 
130 class CSVM;
131 
157 class CKernel : public CSGObject
158 {
169  friend class CDiceKernelNormalizer;
171 
172  friend class CStreamingKernel;
173 
174  public:
175 
179  CKernel();
180 
181 
186  CKernel(int32_t size);
187 
194  CKernel(CFeatures* l, CFeatures* r, int32_t size);
195 
196  virtual ~CKernel();
197 
205  inline float64_t kernel(int32_t idx_a, int32_t idx_b)
206  {
207  REQUIRE(idx_a>=0 && idx_b>=0 && idx_a<num_lhs && idx_b<num_rhs,
208  "%s::kernel(): index out of Range: idx_a=%d/%d idx_b=%d/%d\n",
209  get_name(), idx_a,num_lhs, idx_b,num_rhs);
210 
211  return normalizer->normalize(compute(idx_a, idx_b), idx_a, idx_b);
212  }
213 
219  {
220  return get_kernel_matrix<float64_t>();
221  }
222 
230  preallocated=SGVector<float64_t>())
231  {
232  REQUIRE(lhs, "CKernel::get_kernel_diagonal(): Left-handside "
233  "features missing!\n");
234 
235  REQUIRE(rhs, "CKernel::get_kernel_diagonal(): Right-handside "
236  "features missing!\n");
237 
239  "CKernel::get_kernel_diagonal(): Left- and right-"
240  "handside features must be equal sized\n");
241 
242  /* allocate space if necessary */
243  if (!preallocated.vector)
244  preallocated=SGVector<float64_t>(lhs->get_num_vectors());
245  else
246  {
247  REQUIRE(preallocated.vlen==lhs->get_num_vectors(),
248  "%s::get_kernel_diagonal(): Preallocated vector has"
249  " wrong size!\n", get_name());
250  }
251 
252  for (index_t i=0; i<preallocated.vlen; ++i)
253  preallocated[i]=kernel(i, i);
254 
255  return preallocated;
256  }
257 
264  {
265 
267 
268  for (int32_t i=0; i!=num_rhs; i++)
269  col[i] = kernel(i,j);
270 
271  return col;
272  }
273 
274 
281  {
283 
284  for (int32_t j=0; j!=num_lhs; j++)
285  row[j] = kernel(i,j);
286 
287  return row;
288  }
289 
313  virtual float64_t sum_symmetric_block(index_t block_begin,
314  index_t block_size, bool no_diag=true);
315 
344  virtual float64_t sum_block(index_t block_begin_row,
345  index_t block_begin_col, index_t block_size_row,
346  index_t block_size_col, bool no_diag=false);
347 
372  block_begin, index_t block_size, bool no_diag=true);
373 
404  index_t block_begin, index_t block_size, bool no_diag=true);
405 
442  index_t block_begin_row, index_t block_begin_col,
443  index_t block_size_row, index_t block_size_col,
444  bool no_diag=false);
445 
450  template <class T> SGMatrix<T> get_kernel_matrix();
451 
462  virtual bool init(CFeatures* lhs, CFeatures* rhs);
463 
469 
475 
479  virtual bool init_normalizer();
480 
487  virtual void cleanup();
488 
493  void load(CFile* loader);
494 
499  void save(CFile* writer);
500 
505  inline CFeatures* get_lhs() { SG_REF(lhs); return lhs; }
506 
511  inline CFeatures* get_rhs() { SG_REF(rhs); return rhs; }
512 
517  virtual int32_t get_num_vec_lhs()
518  {
519  return num_lhs;
520  }
521 
526  virtual int32_t get_num_vec_rhs()
527  {
528  return num_rhs;
529  }
530 
535  virtual bool has_features()
536  {
537  return lhs && rhs;
538  }
539 
544  inline bool get_lhs_equals_rhs()
545  {
546  return lhs_equals_rhs;
547  }
548 
550  virtual void remove_lhs_and_rhs();
551 
553  virtual void remove_lhs();
554 
556  virtual void remove_rhs();
557 
565  virtual EKernelType get_kernel_type()=0 ;
566 
573  virtual EFeatureType get_feature_type()=0;
574 
581  virtual EFeatureClass get_feature_class()=0;
582 
587  inline void set_cache_size(int32_t size)
588  {
589  cache_size = size;
590 #ifdef USE_SVMLIGHT
591  cache_reset();
592 #endif //USE_SVMLIGHT
593  }
594 
599  inline int32_t get_cache_size() { return cache_size; }
600 
601 #ifdef USE_SVMLIGHT
602 
604 
609  inline int32_t get_max_elems_cache() { return kernel_cache.max_elems; }
610 
615  inline int32_t get_activenum_cache() { return kernel_cache.activenum; }
616 
624  void get_kernel_row(
625  int32_t docnum, int32_t *active2dnum, float64_t *buffer,
626  bool full_line=false);
627 
632  void cache_kernel_row(int32_t x);
633 
639  void cache_multiple_kernel_rows(int32_t* key, int32_t varnum);
640 
642  void kernel_cache_reset_lru();
643 
650  void kernel_cache_shrink(
651  int32_t totdoc, int32_t num_shrink, int32_t *after);
652 
659  bool regression_hack=false);
660 
665  inline void set_time(int32_t t)
666  {
667  kernel_cache.time=t;
668  }
669 
675  inline int32_t kernel_cache_touch(int32_t cacheidx)
676  {
677  if(kernel_cache.index[cacheidx] != -1)
678  {
679  kernel_cache.lru[kernel_cache.index[cacheidx]]=kernel_cache.time;
680  return(1);
681  }
682  return(0);
683  }
684 
690  inline int32_t kernel_cache_check(int32_t cacheidx)
691  {
692  return(kernel_cache.index[cacheidx] >= 0);
693  }
694 
700  {
701  return(kernel_cache.elems < kernel_cache.max_elems);
702  }
703 
709  void kernel_cache_init(int32_t size, bool regression_hack=false);
710 
712  void kernel_cache_cleanup();
713 
714 #endif //USE_SVMLIGHT
715 
717  void list_kernel();
718 
724  inline bool has_property(EKernelProperty p) { return (properties & p) != 0; }
725 
729  virtual void clear_normal();
730 
736  virtual void add_to_normal(int32_t vector_idx, float64_t weight);
737 
743 
749 
755 
763  virtual bool init_optimization(
764  int32_t count, int32_t *IDX, float64_t *weights);
765 
770  virtual bool delete_optimization();
771 
777  bool init_optimization_svm(CSVM * svm) ;
778 
784  virtual float64_t compute_optimized(int32_t vector_idx);
785 
794  virtual void compute_batch(
795  int32_t num_vec, int32_t* vec_idx, float64_t* target,
796  int32_t num_suppvec, int32_t* IDX, float64_t* alphas,
797  float64_t factor=1.0);
798 
804 
810 
815  virtual int32_t get_num_subkernels();
816 
822  virtual void compute_by_subkernel(
823  int32_t vector_idx, float64_t * subkernel_contrib);
824 
830  virtual const float64_t* get_subkernel_weights(int32_t& num_weights);
831 
837 
842  virtual void set_subkernel_weights(SGVector<float64_t> weights);
843 
852  const TParameter* param, index_t index=-1)
853  {
854  SG_ERROR("Can't compute derivative wrt %s parameter\n", param->m_name)
855  return SGMatrix<float64_t>();
856  }
857 
864  protected:
870  {
871  properties |= p;
872  }
873 
879  {
880  properties &= (properties | p) ^ p;
881  }
882 
887  inline void set_is_initialized(bool p_init) { optimization_initialized=p_init; }
888 
899  virtual float64_t compute(int32_t x, int32_t y)=0;
900 
907  int32_t compute_row_start(int64_t offs, int32_t n, bool symmetric)
908  {
909  int32_t i_start;
910 
911  if (symmetric)
912  i_start=(int32_t) CMath::floor(n-CMath::sqrt(CMath::sq((float64_t) n)-offs));
913  else
914  i_start=(int32_t) (offs/int64_t(n));
915 
916  return i_start;
917  }
918 
923  template <class T> static void* get_kernel_matrix_helper(void* p);
924 
933  virtual void load_serializable_post() throw (ShogunException);
934 
943  virtual void save_serializable_pre() throw (ShogunException);
944 
953  virtual void save_serializable_post() throw (ShogunException);
954 
959  virtual void register_params();
960 
961  private:
964  void init();
965 
966 
967 #ifdef USE_SVMLIGHT
968 #ifndef DOXYGEN_SHOULD_SKIP_THIS
969 
970  struct KERNEL_CACHE {
972  int32_t *index;
974  int32_t *invindex;
976  int32_t *active2totdoc;
978  int32_t *totdoc2active;
980  int32_t *lru;
982  int32_t *occu;
984  int32_t elems;
986  int32_t max_elems;
988  int32_t time;
990  int32_t activenum;
991 
993  KERNELCACHE_ELEM *buffer;
995  KERNELCACHE_IDX buffsize;
996  };
997 
999  struct S_KTHREAD_PARAM
1000  {
1002  CKernel* kernel;
1004  KERNEL_CACHE* kernel_cache;
1006  KERNELCACHE_ELEM** cache;
1008  int32_t* uncached_rows;
1010  int32_t num_uncached;
1012  uint8_t* needs_computation;
1014  int32_t start;
1016  int32_t end;
1018  int32_t num_vectors;
1019  };
1020 #endif // DOXYGEN_SHOULD_SKIP_THIS
1021 
1023  static void* cache_multiple_kernel_row_helper(void* p);
1024 
1026  void kernel_cache_free(int32_t cacheidx);
1027  int32_t kernel_cache_malloc();
1028  int32_t kernel_cache_free_lru();
1029  KERNELCACHE_ELEM *kernel_cache_clean_and_malloc(int32_t cacheidx);
1030 #endif //USE_SVMLIGHT
1031 
1032 
1033  protected:
1035  int32_t cache_size;
1036 
1037 #ifdef USE_SVMLIGHT
1038 
1039  KERNEL_CACHE kernel_cache;
1040 #endif //USE_SVMLIGHT
1041 
1045 
1050 
1053 
1055  int32_t num_lhs;
1057  int32_t num_rhs;
1058 
1061 
1068 
1070  uint64_t properties;
1071 
1075 };
1076 
1077 }
1078 #endif /* _KERNEL_H__ */

SHOGUN Machine Learning Toolbox - Documentation