wolffd@0: /***********************************************************************/
wolffd@0: /*                                                                     */
wolffd@0: /*   svm_learn.h                                                       */
wolffd@0: /*                                                                     */
wolffd@0: /*   Declarations for learning module of Support Vector Machine.       */
wolffd@0: /*                                                                     */
wolffd@0: /*   Author: Thorsten Joachims                                         */
wolffd@0: /*   Date: 02.07.02                                                    */
wolffd@0: /*                                                                     */
wolffd@0: /*   Copyright (c) 2002  Thorsten Joachims - All rights reserved       */
wolffd@0: /*                                                                     */
wolffd@0: /*   This software is available for non-commercial use only. It must   */
wolffd@0: /*   not be modified and distributed without prior permission of the   */
wolffd@0: /*   author. The author is not responsible for implications from the   */
wolffd@0: /*   use of this software.                                             */
wolffd@0: /*                                                                     */
wolffd@0: /***********************************************************************/
wolffd@0: 
wolffd@0: #ifndef SVM_LEARN
wolffd@0: #define SVM_LEARN
wolffd@0: 
wolffd@0: void   svm_learn_classification(DOC **, double *, long, long, LEARN_PARM *, 
wolffd@0: 				KERNEL_PARM *, KERNEL_CACHE *, MODEL *,
wolffd@0: 				double *);
wolffd@0: void   svm_learn_regression(DOC **, double *, long, long, LEARN_PARM *, 
wolffd@0: 			    KERNEL_PARM *, KERNEL_CACHE **, MODEL *);
wolffd@0: void   svm_learn_ranking(DOC **, double *, long, long, LEARN_PARM *, 
wolffd@0: 			 KERNEL_PARM *, KERNEL_CACHE **, MODEL *);
wolffd@0: void   svm_learn_optimization(DOC **, double *, long, long, LEARN_PARM *, 
wolffd@0: 			      KERNEL_PARM *, KERNEL_CACHE *, MODEL *,
wolffd@0: 			      double *);
wolffd@0: long   optimize_to_convergence(DOC **, long *, long, long, LEARN_PARM *,
wolffd@0: 			       KERNEL_PARM *, KERNEL_CACHE *, SHRINK_STATE *,
wolffd@0: 			       MODEL *, long *, long *, double *,
wolffd@0: 			       double *, double *,
wolffd@0: 			       TIMING *, double *, long, long);
wolffd@0: long   optimize_to_convergence_sharedslack(DOC **, long *, long, long, 
wolffd@0: 			       LEARN_PARM *,
wolffd@0: 			       KERNEL_PARM *, KERNEL_CACHE *, SHRINK_STATE *,
wolffd@0: 			       MODEL *, double *, double *, double *,
wolffd@0: 			       TIMING *, double *);
wolffd@0: double compute_objective_function(double *, double *, double *, double,
wolffd@0: 				  long *, long *);
wolffd@0: void   clear_index(long *);
wolffd@0: void   add_to_index(long *, long);
wolffd@0: long   compute_index(long *,long, long *);
wolffd@0: void   optimize_svm(DOC **, long *, long *, long *, double, long *, long *, 
wolffd@0: 		    MODEL *, 
wolffd@0: 		    long, long *, long, double *, double *, double *, 
wolffd@0: 		    LEARN_PARM *, CFLOAT *, KERNEL_PARM *, QP *, double *);
wolffd@0: void   compute_matrices_for_optimization(DOC **, long *, long *, long *, double,
wolffd@0: 					 long *,
wolffd@0: 					 long *, long *, MODEL *, double *, 
wolffd@0: 					 double *, double *, long, long, LEARN_PARM *, 
wolffd@0: 					 CFLOAT *, KERNEL_PARM *, QP *);
wolffd@0: long   calculate_svm_model(DOC **, long *, long *, double *, double *, 
wolffd@0: 			   double *, double *, LEARN_PARM *, long *,
wolffd@0: 			   long *, MODEL *);
wolffd@0: long   check_optimality(MODEL *, long *, long *, double *, double *,
wolffd@0: 			double *, long, 
wolffd@0: 			LEARN_PARM *,double *, double, long *, long *, long *,
wolffd@0: 			long *, long, KERNEL_PARM *);
wolffd@0: long   check_optimality_sharedslack(DOC **docs, MODEL *model, long int *label,
wolffd@0: 		      double *a, double *lin, double *c, double *slack,
wolffd@0: 		      double *alphaslack, long int totdoc, 
wolffd@0: 		      LEARN_PARM *learn_parm, double *maxdiff, 
wolffd@0: 		      double epsilon_crit_org, long int *misclassified, 
wolffd@0: 		      long int *active2dnum,
wolffd@0: 		      long int *last_suboptimal_at, 
wolffd@0: 		      long int iteration, KERNEL_PARM *kernel_parm);
wolffd@0: void   compute_shared_slacks(DOC **docs, long int *label, double *a, 
wolffd@0: 			     double *lin, double *c, long int *active2dnum, 
wolffd@0: 			     LEARN_PARM *learn_parm,
wolffd@0: 			     double *slack, double *alphaslack);
wolffd@0: long   identify_inconsistent(double *, long *, long *, long, LEARN_PARM *, 
wolffd@0: 			     long *, long *);
wolffd@0: long   identify_misclassified(double *, long *, long *, long,
wolffd@0: 			      MODEL *, long *, long *);
wolffd@0: long   identify_one_misclassified(double *, long *, long *, long,
wolffd@0: 				  MODEL *, long *, long *);
wolffd@0: long   incorporate_unlabeled_examples(MODEL *, long *,long *, long *,
wolffd@0: 				      double *, double *, long, double *,
wolffd@0: 				      long *, long *, long, KERNEL_PARM *,
wolffd@0: 				      LEARN_PARM *);
wolffd@0: void   update_linear_component(DOC **, long *, long *, double *, double *, 
wolffd@0: 			       long *, long, long, KERNEL_PARM *, 
wolffd@0: 			       KERNEL_CACHE *, double *,
wolffd@0: 			       CFLOAT *, double *);
wolffd@0: long   select_next_qp_subproblem_grad(long *, long *, double *, 
wolffd@0: 				      double *, double *, long,
wolffd@0: 				      long, LEARN_PARM *, long *, long *, 
wolffd@0: 				      long *, double *, long *, KERNEL_CACHE *,
wolffd@0: 				      long, long *, long *);
wolffd@0: long   select_next_qp_subproblem_rand(long *, long *, double *, 
wolffd@0: 				      double *, double *, long,
wolffd@0: 				      long, LEARN_PARM *, long *, long *, 
wolffd@0: 				      long *, double *, long *, KERNEL_CACHE *,
wolffd@0: 				      long *, long *, long);
wolffd@0: long   select_next_qp_slackset(DOC **docs, long int *label, double *a, 
wolffd@0: 			       double *lin, double *slack, double *alphaslack, 
wolffd@0: 			       double *c, LEARN_PARM *learn_parm, 
wolffd@0: 			       long int *active2dnum, double *maxviol);
wolffd@0: void   select_top_n(double *, long, long *, long);
wolffd@0: void   init_shrink_state(SHRINK_STATE *, long, long);
wolffd@0: void   shrink_state_cleanup(SHRINK_STATE *);
wolffd@0: long   shrink_problem(DOC **, LEARN_PARM *, SHRINK_STATE *, KERNEL_PARM *, 
wolffd@0: 		      long *, long *, long, long, long, double *, long *);
wolffd@0: void   reactivate_inactive_examples(long *, long *, double *, SHRINK_STATE *,
wolffd@0: 				    double *, double*, long, long, long, LEARN_PARM *, 
wolffd@0: 				    long *, DOC **, KERNEL_PARM *,
wolffd@0: 				    KERNEL_CACHE *, MODEL *, CFLOAT *, 
wolffd@0: 				    double *, double *);
wolffd@0: 
wolffd@0: /* cache kernel evalutations to improve speed */
wolffd@0: KERNEL_CACHE *kernel_cache_init(long, long);
wolffd@0: void   kernel_cache_cleanup(KERNEL_CACHE *);
wolffd@0: void   get_kernel_row(KERNEL_CACHE *,DOC **, long, long, long *, CFLOAT *, 
wolffd@0: 		      KERNEL_PARM *);
wolffd@0: void   cache_kernel_row(KERNEL_CACHE *,DOC **, long, KERNEL_PARM *);
wolffd@0: void   cache_multiple_kernel_rows(KERNEL_CACHE *,DOC **, long *, long, 
wolffd@0: 				  KERNEL_PARM *);
wolffd@0: void   kernel_cache_shrink(KERNEL_CACHE *,long, long, long *);
wolffd@0: void   kernel_cache_reset_lru(KERNEL_CACHE *);
wolffd@0: long   kernel_cache_malloc(KERNEL_CACHE *);
wolffd@0: void   kernel_cache_free(KERNEL_CACHE *,long);
wolffd@0: long   kernel_cache_free_lru(KERNEL_CACHE *);
wolffd@0: CFLOAT *kernel_cache_clean_and_malloc(KERNEL_CACHE *,long);
wolffd@0: long   kernel_cache_touch(KERNEL_CACHE *,long);
wolffd@0: long   kernel_cache_check(KERNEL_CACHE *,long);
wolffd@0: long   kernel_cache_space_available(KERNEL_CACHE *);
wolffd@0: 
wolffd@0: void compute_xa_estimates(MODEL *, long *, long *, long, DOC **, 
wolffd@0: 			  double *, double *, KERNEL_PARM *, 
wolffd@0: 			  LEARN_PARM *, double *, double *, double *);
wolffd@0: double xa_estimate_error(MODEL *, long *, long *, long, DOC **, 
wolffd@0: 			 double *, double *, KERNEL_PARM *, 
wolffd@0: 			 LEARN_PARM *);
wolffd@0: double xa_estimate_recall(MODEL *, long *, long *, long, DOC **, 
wolffd@0: 			  double *, double *, KERNEL_PARM *, 
wolffd@0: 			  LEARN_PARM *);
wolffd@0: double xa_estimate_precision(MODEL *, long *, long *, long, DOC **, 
wolffd@0: 			     double *, double *, KERNEL_PARM *, 
wolffd@0: 			     LEARN_PARM *);
wolffd@0: void avg_similarity_of_sv_of_one_class(MODEL *, DOC **, double *, long *, KERNEL_PARM *, double *, double *);
wolffd@0: double most_similar_sv_of_same_class(MODEL *, DOC **, double *, long, long *, KERNEL_PARM *, LEARN_PARM *);
wolffd@0: double distribute_alpha_t_greedily(long *, long, DOC **, double *, long, long *, KERNEL_PARM *, LEARN_PARM *, double);
wolffd@0: double distribute_alpha_t_greedily_noindex(MODEL *, DOC **, double *, long, long *, KERNEL_PARM *, LEARN_PARM *, double); 
wolffd@0: void estimate_transduction_quality(MODEL *, long *, long *, long, DOC **, double *);
wolffd@0: double estimate_margin_vcdim(MODEL *, double, double, KERNEL_PARM *);
wolffd@0: double estimate_sphere(MODEL *, KERNEL_PARM *);
wolffd@0: double estimate_r_delta_average(DOC **, long, KERNEL_PARM *); 
wolffd@0: double estimate_r_delta(DOC **, long, KERNEL_PARM *); 
wolffd@0: double length_of_longest_document_vector(DOC **, long, KERNEL_PARM *); 
wolffd@0: 
wolffd@0: void   write_model(char *, MODEL *);
wolffd@0: void   write_prediction(char *, MODEL *, double *, double *, long *, long *,
wolffd@0: 			long, LEARN_PARM *);
wolffd@0: void   write_alphas(char *, double *, long *, long);
wolffd@0: 
wolffd@0: typedef struct cache_parm_s {
wolffd@0:   KERNEL_CACHE *kernel_cache;
wolffd@0:   CFLOAT *cache;
wolffd@0:   DOC **docs; 
wolffd@0:   long m;
wolffd@0:   KERNEL_PARM *kernel_parm;
wolffd@0:   long offset,stepsize;
wolffd@0: } cache_parm_t;
wolffd@0: 
wolffd@0: #endif