/************************************************************
 * HMMER - Biological sequence analysis with profile-HMMs
 * Copyright (C) 1992-1998 Sean R. Eddy
 *
 *   This source code is distributed under the terms of the
 *   GNU General Public License. See the files COPYING and
 *   GNULICENSE for details.
 *
 ************************************************************/

/* hmmbuild.c
 * SRE, Mon Nov 18 12:41:29 1996
 *
 * main() for HMM construction from an alignment.
 * RCS $Id: hmmbuild.c,v 1.1.1.1 2001/06/18 13:59:49 birney Exp $
 */

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "structs.h"		/* data structures, macros, #define's   */
#include "config.h"		/* compile-time configuration constants */
#include "funcs.h"		/* function declarations                */
#include "globals.h"		/* alphabet global variables            */
#include "squid.h"		/* general sequence analysis library    */

#ifdef MEMDEBUG
#include "dbmalloc.h"
#endif

static char banner[] = "hmmbuild - build a hidden Markov model from an alignment";

static char usage[]  = "\
Usage: hmmbuild [-options] <hmmfile output> <alignment file>\n\
  Available options are:\n\
   -h        : help; print brief help on version and usage\n\
   -n <s>    : name; name this HMM <s>\n\
   -o <file> : re-save annotated alignment to <file>\n\
   -A        : append; append this HMM to <hmmfile>\n\
   -F        : force; allow overwriting of <hmmfile>\n\
\n\
  Alternative search algorithm styles: (default: hmmls domain alignment)\n\
   -f        : multi-hit local (hmmfs style)\n\
   -g        : global alignment (hmms style, Needleman/Wunsch)\n\
   -s        : local alignment (hmmsw style, Smith/Waterman)\n\
";

static char experts[] = "\
  Alternative model construction strategies: (default: MAP)\n\
   --fast    : Krogh/Haussler fast heuristic construction (see --gapmax)\n\
   --hand    : manual construction (requires SELEX file, #=RF annotation)\n\
\n\
  Expert customization of parameters and priors:\n\
   --null  <file> : read null (random sequence) model from <file>\n\
   --pam <file>   : heuristic PAM-based prior, using BLAST PAM matrix in <file>\n\
   --prior <file> : read Dirichlet prior parameters from <file>\n\
\n\
  Alternative sequence weighting strategies: (default: GSC weights)\n\
   --wblosum     : Henikoff simple filter weights (see --idlevel)\n\
   --wgsc        : Gerstein/Sonnhammer/Chothia tree weights (default)\n\
   --wme         : maximum entropy (ME)\n\
   --wvoronoi    : Sibbald/Argos Voronoi weights\n\
   --wnone       : don't do any weighting\n\
   --noeff       : don't use effective sequence number; just use nseq\n\
\n\
  Forcing an alphabet: (normally autodetected)\n\
   --amino   : override autodetection, assert that seqs are protein\n\
   --nucleic : override autodetection, assert that seqs are DNA/RNA\n\
\n\
  Other expert options:\n\
   --archpri <x> : set architecture size prior to <x> {0.85} [0..1]\n\
   --binary      : save the model in binary format, not ASCII text\n\
   --cfile <file>: save count vectors to <file>\n\
   --gapmax <x>  : max fraction of gaps in mat column {0.50} [0..1]\n\
   --idlevel <x> : set frac. id level used by eff. nseq and --wblosum {0.62}\n\
   --pamwgt <x>  : set weight on PAM-based prior to <x> {20.}[>=0]\n\
   --swentry <x> : set S/W aggregate entry prob. to <x> {0.5}\n\
   --swexit <x>  : set S/W aggregate exit prob. to <x>  {0.5}\n\
   --verbose     : print a lot of boring information\n\
\n";

static struct opt_s OPTIONS[] = {
  { "-f", TRUE, sqdARG_NONE },
  { "-g", TRUE, sqdARG_NONE }, 
  { "-h", TRUE, sqdARG_NONE }, 
  { "-n", TRUE, sqdARG_STRING},  
  { "-o", TRUE, sqdARG_STRING},
  { "-s", TRUE, sqdARG_NONE }, 
  { "-A", TRUE, sqdARG_NONE },
  { "-F", TRUE, sqdARG_NONE },
  { "--amino",   FALSE, sqdARG_NONE  },
  { "--archpri", FALSE, sqdARG_FLOAT }, 
  { "--binary",  FALSE, sqdARG_NONE  }, 
  { "--cfile",   FALSE, sqdARG_STRING},
  { "--fast",    FALSE, sqdARG_NONE},
  { "--gapmax",  FALSE, sqdARG_FLOAT },
  { "--hand",    FALSE, sqdARG_NONE},
  { "--idlevel", FALSE, sqdARG_FLOAT },
  { "--noeff",   FALSE, sqdARG_NONE },
  { "--nucleic", FALSE, sqdARG_NONE },
  { "--null",    FALSE, sqdARG_STRING },
  { "--pam",     FALSE, sqdARG_STRING },
  { "--pamwgt",  FALSE, sqdARG_FLOAT },
  { "--prior",   FALSE, sqdARG_STRING },
  { "--swentry", FALSE, sqdARG_FLOAT },
  { "--swexit",  FALSE, sqdARG_FLOAT },
  { "--verbose", FALSE, sqdARG_NONE  },
  { "--wgsc",    FALSE, sqdARG_NONE },
  { "--wblosum", FALSE, sqdARG_NONE },
  { "--wme",     FALSE, sqdARG_NONE },
  { "--wnone",   FALSE, sqdARG_NONE },
  { "--wvoronoi",FALSE, sqdARG_NONE },
};
#define NOPTIONS (sizeof(OPTIONS) / sizeof(struct opt_s))

static void save_model(struct plan7_s *hmm, char *hmmfile, int do_append, int do_binary);
static void print_all_scores(FILE *fp, struct plan7_s *hmm, 
			     AINFO *ainfo, char **dsq, int nseq, 
			     struct p7trace_s **tr);
static void save_countvectors(char *cfile, struct plan7_s *hmm);
static void position_average_score(struct plan7_s *hmm, char **seq, float *wgt,
				   int nseq, struct p7trace_s **tr, float *pernode,
				   float *ret_avg);
static float frag_trace_score(struct plan7_s *hmm, char *dsq, struct p7trace_s *tr, 
			      float *pernode, float expected);
static void maximum_entropy(struct plan7_s *hmm, char **dsq, AINFO *ainfo, 
			    int nseq, float eff_nseq, 
			    struct p7prior_s *prior, struct p7trace_s **tr);


int
main(int argc, char **argv) 
{
  char            *seqfile;     /* seqfile to read alignment from          */ 
  int              format;	/* format of seqfile                       */
  char           **aseq;        /* aligned sequences. [0.nseq-1][0.alen-1] */
  AINFO            ainfo;	/* optional info attached to aseq's        */
  char           **dsq;         /* digitized unaligned aseq's              */ 
  struct plan7_s  *hmm;         /* constructed HMM; written to hmmfile     */
  struct p7prior_s *pri;        /* Dirichlet priors to use                 */
  struct p7trace_s **tr;        /* fake tracebacks for aseq's              */ 
  char            *hmmfile;     /* file to write HMM to                    */
  FILE            *fp;          /* OUTPUT file handle (misc.)              */
  char            *name;        /* name of the HMM                         */
  int              idx;		/* counter for sequences                   */
  float  randomseq[MAXABET];	/* null sequence model                     */
  float            p1;		/* null sequence model p1 transition       */

  char *optname;                /* name of option found by Getopt()      */
  char *optarg;                 /* argument found by Getopt()            */
  int   optind;                 /* index in argv[]                       */
  enum p7_construction c_strategy; /* construction strategy choice        */
  enum p7_weight {		/* weighting strategy */
    WGT_NONE, WGT_GSC, WGT_BLOSUM, WGT_VORONOI, WGT_ME} w_strategy;
  enum p7_config {              /* algorithm configuration strategy      */
    P7_BASE_CONFIG, P7_LS_CONFIG, P7_FS_CONFIG, P7_SW_CONFIG } cfg_strategy;
  float gapmax;			/* max frac gaps in mat col for -k       */
  int   overwrite_protect;	/* TRUE to prevent overwriting HMM file  */
  int   verbose;		/* TRUE to show a lot of output          */
  char *align_ofile;            /* name of output alignment file         */
  char *rndfile;		/* random sequence model file to read    */
  char *prifile;		/* Dirichlet prior file to read          */
  char *pamfile;		/* PAM matrix file for heuristic prior   */
  char *cfile;			/* output file for count vectors         */
  float archpri;		/* "architecture" prior on model size    */
  float pamwgt;			/* weight on PAM for heuristic prior     */
  int   do_append;		/* TRUE to append to hmmfile             */
  int   do_binary;		/* TRUE to write in binary format        */
  float blosumlevel;		/* BLOSUM frac id filtering level [0.62] */
  float swentry;		/* S/W aggregate entry probability       */
  float swexit;			/* S/W aggregate exit probability        */
  int   do_eff;			/* TRUE to set an effective seq number   */
  float eff_nseq;		/* effective sequence number             */

#ifdef MEMDEBUG
  unsigned long histid1, histid2;
  size_t orig_size, current_size;
  orig_size = malloc_inuse(&histid1);
  fprintf(stderr, "[... memory debugging is ON ...]\n");
#endif

  /*********************************************** 
   * Parse command line
   ***********************************************/
  
  c_strategy        = P7_MAP_CONSTRUCTION;
  w_strategy        = WGT_GSC;
  blosumlevel       = 0.62;
  cfg_strategy      = P7_LS_CONFIG;
  gapmax            = 0.5;
  overwrite_protect = TRUE;
  verbose           = FALSE;
  align_ofile       = NULL;
  rndfile           = NULL;
  prifile           = NULL;
  pamfile           = NULL;
  cfile             = NULL;
  archpri           = 0.85;
  pamwgt            = 20.;
  Alphabet_type     = hmmNOTSETYET;	/* initially unknown */
  name              = NULL;
  do_append         = FALSE; 
  swentry           = 0.5;
  swexit            = 0.5;
  do_eff            = TRUE;
  do_binary         = FALSE;
  
  while (Getopt(argc, argv, OPTIONS, NOPTIONS, usage,
                &optind, &optname, &optarg))  {
    if      (strcmp(optname, "-f") == 0) cfg_strategy      = P7_FS_CONFIG;
    else if (strcmp(optname, "-g") == 0) cfg_strategy      = P7_BASE_CONFIG;
    else if (strcmp(optname, "-n") == 0) name              = Strdup(optarg); 
    else if (strcmp(optname, "-o") == 0) align_ofile       = optarg;
    else if (strcmp(optname, "-r") == 0) rndfile           = optarg;
    else if (strcmp(optname, "-s") == 0) cfg_strategy      = P7_SW_CONFIG;
    if      (strcmp(optname, "-A") == 0) do_append         = TRUE; 
    else if (strcmp(optname, "-F") == 0) overwrite_protect = FALSE;
    else if (strcmp(optname, "--amino")   == 0) SetAlphabet(hmmAMINO);
    else if (strcmp(optname, "--archpri") == 0) archpri       = atof(optarg);
    else if (strcmp(optname, "--binary")  == 0) do_binary     = TRUE;
    else if (strcmp(optname, "--cfile")   == 0) cfile         = optarg;
    else if (strcmp(optname, "--fast")    == 0) c_strategy    = P7_FAST_CONSTRUCTION;
    else if (strcmp(optname, "--hand")    == 0) c_strategy    = P7_HAND_CONSTRUCTION;
    else if (strcmp(optname, "--gapmax")  == 0) gapmax        = atof(optarg);
    else if (strcmp(optname, "--idlevel") == 0) blosumlevel   = atof(optarg);
    else if (strcmp(optname, "--noeff")   == 0) do_eff        = FALSE;
    else if (strcmp(optname, "--nucleic") == 0) SetAlphabet(hmmNUCLEIC);
    else if (strcmp(optname, "--pam")     == 0) pamfile       = optarg;
    else if (strcmp(optname, "--pamwgt")  == 0) pamwgt        = atof(optarg);
    else if (strcmp(optname, "--prior")   == 0) prifile       = optarg;
    else if (strcmp(optname, "--swentry") == 0) swentry       = atof(optarg); 
    else if (strcmp(optname, "--swexit")  == 0) swexit        = atof(optarg); 
    else if (strcmp(optname, "--verbose") == 0) verbose       = TRUE;
    else if (strcmp(optname, "--wgsc")    == 0) w_strategy    = WGT_GSC;
    else if (strcmp(optname, "--wblosum") == 0) w_strategy    = WGT_BLOSUM; 
    else if (strcmp(optname, "--wme")     == 0) w_strategy    = WGT_ME;  
    else if (strcmp(optname, "--wnone")   == 0) w_strategy    = WGT_NONE; 
    else if (strcmp(optname, "--wvoronoi")== 0) w_strategy    = WGT_VORONOI;
    else if (strcmp(optname, "-h") == 0) {
      Banner(stdout, banner);
      puts(usage);
      puts(experts);
      exit(0);
    }
  }
  if (argc - optind != 2)
    Die("Incorrect number of arguments.\n%s\n", usage);

  hmmfile = argv[optind++];
  seqfile = argv[optind++]; 

  if (gapmax < 0. || gapmax > 1.) 
    Die("--gapmax must be a value from 0 to 1\n%s\n", usage);
  if (archpri < 0. || archpri > 1.)
    Die("--archpri must be a value from 0 to 1\n%s\n", usage);
  if (overwrite_protect && !do_append && FileExists(hmmfile))
    Die("HMM file %s already exists. Rename or delete it.", hmmfile); 
  if (overwrite_protect && align_ofile != NULL && FileExists(align_ofile))
    Die("Alignment resave file %s exists. Rename or delete it.", align_ofile); 

  /*********************************************** 
   * Get sequence data
   ***********************************************/

  if (! SeqfileFormat(seqfile, &format, NULL))
    switch (squid_errno) {
    case SQERR_NOFILE: 
      Die("Alignment file %s could not be opened for reading", seqfile);
      /*FALLTHRU*/ /* a white lie to shut lint up */
    case SQERR_FORMAT: 
    default:           
      Die("Failed to determine format of alignment file %s", seqfile);
    }
  
                                /* read the alignment from file */
  if (! ReadAlignment(seqfile, format, &aseq, &ainfo))
    Die("Failed to read aligned sequence file %s", seqfile);
  for (idx = 0; idx < ainfo.nseq; idx++)
    s2upper(aseq[idx]);
				/* Set up the alphabet globals */
  if (Alphabet_type == 0) DetermineAlphabet(aseq, ainfo.nseq);

				/* Set up Dirichlet priors */
  if (prifile == NULL)  pri = P7DefaultPrior();
  else                  pri = P7ReadPrior(prifile);

  if (pamfile != NULL)  PAMPrior(pamfile, pri, pamwgt);

				/* Set up the null/random seq model */
  if (rndfile == NULL)  P7DefaultNullModel(randomseq, &p1);
  else                  P7ReadNullModel(rndfile, randomseq, &p1);

				/* Prepare sequences for internal use */
  DigitizeAlignment(aseq, &ainfo, &dsq);
  
				/* In some respects we treat DNA more crudely... */
  if (Alphabet_type == hmmNUCLEIC)
    {
      do_eff     = FALSE;	/* don't do effective seq #; it's calibrated for protein */
    }

  /*********************************************** 
   * Show the banner
   ***********************************************/

  Banner(stdout, banner);
  printf("Training alignment:                %s\n", seqfile);
  printf("Number of sequences:               %d\n", ainfo.nseq);

  printf("Search algorithm configuration:    ");
  if      (cfg_strategy == P7_BASE_CONFIG)   puts("Global alignment (hmms)");
  else if (cfg_strategy == P7_SW_CONFIG)     {
    puts("Local  (hmmsw)");
    printf("S/W aggregate entry probability:   %.2f\n", swentry);
    printf("S/W aggregate exit probability:    %.2f\n", swexit);
  }
  else if (cfg_strategy == P7_LS_CONFIG)     puts("Multiple domain (hmmls)");
  else if (cfg_strategy == P7_FS_CONFIG)     {
    puts("Multiple local (hmmfs)");
    printf("S/W aggregate entry probability:   %.2f\n", swentry);
    printf("S/W aggregate exit probability:    %.2f\n", swexit);
  }

  printf("Model construction strategy:       ");
  if (c_strategy == P7_HAND_CONSTRUCTION)    puts("Manual, from #=RF annotation");
  else if (c_strategy==P7_FAST_CONSTRUCTION) printf("Fast/ad hoc (gapmax %.2f)\n", gapmax);
  else                                       printf("MAP (gapmax hint: %.2f)\n", gapmax);

  printf("Prior used:                        %s\n",
	 (prifile == NULL) ? "(default)" : prifile);

  printf("Prior strategy:                    ");
  if      (pri->strategy == PRI_DCHLET) puts("Dirichlet");
  else if (pri->strategy == PRI_PAM)    puts("PAM hack");
  if (pri->strategy == PRI_PAM)
    printf("PAM prior weight:                  %.1f\n", pamwgt);

  printf("Sequence weighting method:         ");
  if      (w_strategy == WGT_NONE)   puts("none");
  else if (w_strategy == WGT_GSC)    puts("G/S/C tree weights");
  else if (w_strategy == WGT_BLOSUM) printf("BLOSUM filter at %.2f id\n", blosumlevel);
  else if (w_strategy == WGT_VORONOI)puts("Sibbald/Argos Voronoi");
  else if (w_strategy == WGT_ME)     puts("Maximum entropy");
  printf("- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -\n");

  /*********************************************** 
   * Build an HMM
   ***********************************************/

  /* Determine the effective sequence number to use (optional)
   */
  eff_nseq = (float) ainfo.nseq;
  if (do_eff)
    {
      float *wgt;
      printf("%-40s ... ", "Determining effective sequence number");
      fflush(stdout);
				/* protect weights */
      wgt = MallocOrDie(sizeof(float) * ainfo.nseq);
      FCopy(wgt, ainfo.wgt, ainfo.nseq);
				/* use BlosumWeights for now... */
      BlosumWeights(aseq, &ainfo, blosumlevel);
      eff_nseq = FSum(ainfo.wgt, ainfo.nseq);
				/* re-install old weights */
      FCopy(ainfo.wgt, wgt, ainfo.nseq);
      free(wgt);
      printf("done. [%.0f]\n", eff_nseq);
    }


  /* Weight the sequences (optional),
   */
  if (w_strategy == WGT_GSC || w_strategy == WGT_BLOSUM || w_strategy == WGT_VORONOI)
    {
      printf("%-40s ... ", "Weighting sequences heuristically");
      fflush(stdout);
    }

  if      (w_strategy == WGT_GSC)     GSCWeights(aseq, &ainfo);
  else if (w_strategy == WGT_BLOSUM)  BlosumWeights(aseq, &ainfo, blosumlevel);
  else if (w_strategy == WGT_VORONOI) VoronoiWeights(aseq, &ainfo); 

  if (w_strategy == WGT_GSC || w_strategy == WGT_BLOSUM || w_strategy == WGT_VORONOI)
    printf("done.\n");


  /* Set the effective sequence number (if do_eff is FALSE, eff_nseq 
   * was set to nseq).
   */
  FNorm(ainfo.wgt, ainfo.nseq);
  FScale(ainfo.wgt, ainfo.nseq, eff_nseq);


  /* Build a model architecture.
   * If we're not doing MD or ME, that's all we need to do.
   * We get an allocated, counts-based HMM back.
   */
  printf("%-40s ... ", "Constructing model architecture");
  fflush(stdout);
  if (c_strategy == P7_FAST_CONSTRUCTION)
    P7Fastmodelmaker(aseq, dsq, &ainfo, gapmax, &hmm, &tr);
  else if (c_strategy == P7_HAND_CONSTRUCTION)
    P7Handmodelmaker(aseq, dsq, &ainfo, &hmm, &tr);
  else
    P7Maxmodelmaker(aseq, dsq, &ainfo, gapmax, 
		    pri, randomseq, p1, archpri, &hmm, &tr);
  printf("done.\n");

  /* Save the count vectors if asked. Used primarily for
   * making the data files for training priors.
   */
  if (cfile != NULL) 
    {
      printf("%-40s ... ", "Saving count vector file");
      fflush(stdout);
      save_countvectors(cfile, hmm); 
      printf("done. [%s]\n", cfile);
    }

  /* Record the null model in the HMM;
   * add prior contributions in pseudocounts and renormalize.
   */
  printf("%-40s ... ", "Converting counts to probabilities");
  fflush(stdout);
  Plan7SetNullModel(hmm, randomseq, p1);
  P7PriorifyHMM(hmm, pri);
  printf("done.\n");


  /* Model configuration, temporary.
   * hmmbuild assumes that it's given an alignment of single domains,
   * and the alignment may contain fragments. So, for the purpose of
   * scoring the sequences (or, optionally, MD/ME weighting),
   * configure the model into hmmsw mode. Later we'll
   * configure the model according to how the user wants to
   * use it.
   */
  Plan7SWConfig(hmm, 0.5, 0.5);

  /* Do model-dependent "weighting" strategies.
   */
  if (w_strategy == WGT_ME)
    {
      printf("\n%-40s ...\n", "Maximum entropy weighting, iterative");
      maximum_entropy(hmm, dsq, &ainfo, ainfo.nseq, eff_nseq, pri, tr);
      printf("----------------------------------------------\n\n");
    }

  /* Give the model a name; by default, the name of the alignment file
   * without any filename extension (i.e. "globins.slx" becomes "globins"
   */
  printf("%-40s ... ", "Setting model name, etc.");
  fflush(stdout);
  if (name == NULL) name = FileTail(seqfile, TRUE);
  Plan7SetName(hmm, name);
  Plan7ComlogAppend(hmm, argc, argv);
  Plan7SetCtime(hmm);
  hmm->nseq = ainfo.nseq;
  printf("done. [%s]\n", name); 
  free(name);
   
  /* Print information for the user
   */
  printf("\nConstructed a profile HMM (length %d)\n", hmm->M);
  PrintPlan7Stats(stdout, hmm, dsq, ainfo.nseq, tr); 
  printf("\n");

  /* Configure the model for chosen algorithm
   */
  printf("%-40s ... ", "Finalizing model configuration");
  fflush(stdout);
  switch (cfg_strategy) {
  case P7_BASE_CONFIG:  Plan7GlobalConfig(hmm);              break;
  case P7_SW_CONFIG:    Plan7SWConfig(hmm, swentry, swexit); break;
  case P7_LS_CONFIG:    Plan7LSConfig(hmm);                  break;
  case P7_FS_CONFIG:    Plan7FSConfig(hmm, swentry, swexit); break;
  default:              Die("bogus configuration choice");
  }
  printf("done.\n");

  /* Save new HMM to disk: open a file for appending or writing.
   */
  printf("%-40s ... ", "Saving model to file");
  fflush(stdout);
  save_model(hmm, hmmfile, do_append, do_binary);
  printf("done. [%s]\n", hmmfile);

				/* the annotated alignment may be resaved */
  if (align_ofile != NULL) {
    char  **new_aseq;
    AINFO   new_ainfo;

    printf("%-40s ... ", "Saving annotated alignment");
    fflush(stdout);
    P7Traces2Alignment(dsq, ainfo.sqinfo, ainfo.wgt, ainfo.nseq, hmm->M, tr, FALSE,
		       &new_aseq, &new_ainfo);
    if ((fp = fopen(align_ofile, "w")) == NULL) {
      Warn("Failed to open alignment resave file %s; using stdout instead", 
	   align_ofile);
      fp = stdout;
    }
    WriteSELEX(fp, new_aseq, &new_ainfo, 50);
    if (fp != stdout) fclose(fp);
    FreeAlignment(new_aseq, &new_ainfo);
    printf("done. [%s]\n", align_ofile);
  }
  
  /* Verbose output; show scores for each sequence
   */
  if (verbose)
    print_all_scores(stdout, hmm, &ainfo, dsq, ainfo.nseq, tr);

  /* Clean up and exit
   */
  for (idx = 0; idx < ainfo.nseq; idx++) P7FreeTrace(tr[idx]);
  free(tr);
  FreePlan7(hmm);
  P7FreePrior(pri);
  FreeAlignment(aseq, &ainfo);
  Free2DArray(dsq, ainfo.nseq); 
  SqdClean();

#ifdef MEMDEBUG
  current_size = malloc_inuse(&histid2);
  if (current_size != orig_size) {
    printf("** I have %d bytes still allocated, pal\n",
	   current_size- orig_size);
    malloc_list(2, histid1, histid2);
  }
  else fprintf(stderr, "[No memory leaks.]\n");
#endif
  return 0;
}

/* Function: save_model()
 * 
 * Purpose:  Save the new model to a file.
 * 
 * Args:     hmm       - model to save
 *           hmmfile   - file to save to (if NULL, use stdout)      
 *           do_append - TRUE to append to file
 *           do_binary - TRUE to write a binary file
 *           
 * Return:   (void)
 */          
static void
save_model(struct plan7_s *hmm, char *hmmfile, int do_append, int do_binary)
{
  FILE    *fp;

  if (hmmfile == NULL)
    fp = stdout;
  else if (do_append)
    {
				/* check that it looks like an HMM file */
#ifdef REMOVED			/* This code induces an unresolved Linux/SGI NFS bug! */
      if (FileExists(hmmfile)) 
	{ 
	  HMMFILE *hmmfp;
	  hmmfp = HMMFileOpen(hmmfile, NULL);
	  if (hmmfp == NULL) {
	    Warn("%s not an HMM file; can't append to it; using stdout instead",
		 hmmfile);
	    fp = stdout;
	    puts("");		/* do a newline before stdout HMM starts */
	  } else {
	    HMMFileClose(hmmfp);
	  }
	}
#endif

      if ((fp = fopen(hmmfile, "a")) == NULL) {
	Warn("hey, where'd your HMM file go? Using stdout instead.");
	fp = stdout;
	puts("");		/* do a newline before stdout HMM starts */
      } 
    } 
  else 
    {
      if ((fp = fopen(hmmfile, "w")) == NULL) {
	Warn("Failed to open HMM save file %s; using stdout instead", hmmfile);
	fp = stdout;
	puts("");		/* do a newline before stdout HMM starts */
      }
    }

  if (do_binary) WriteBinHMM(fp, hmm);
  else           WriteAscHMM(fp, hmm);

  if (fp != stdout) fclose(fp);
  return;
}





/* Function: print_all_scores()
 * 
 * Purpose:  For each training sequence, print its score under
 *           the final model.
 *           
 * Args:     fp   - where to print the output (usu. stdout)
 *           hmm  - newly constructed HMM, with prob's.
 *           ainfo- info with aseq
 *           dsq  - digitized unaligned training sequences.
 *           nseq - number of training sequences
 *           tr   - array of tracebacks
 *           
 * Return:   (void)                         
 */
static void
print_all_scores(FILE *fp, struct plan7_s *hmm,
		 AINFO *ainfo, char **dsq, int nseq, struct p7trace_s **tr)
{
  int idx;			/* counter for sequences */

				/* make sure model scores are ready */
  P7Logoddsify(hmm, TRUE);
				/* header */
  fputs("**\n", fp);
  fputs("Individual training sequence scores:\n", fp);
				/* score for each sequence */
  for (idx = 0; idx < nseq; idx++) 
    {
      fprintf(fp, "%7.2f %-12s %s\n", 
	      P7TraceScore(hmm, dsq[idx], tr[idx]),
	      ainfo->sqinfo[idx].name,
	      (ainfo->sqinfo[idx].flags & SQINFO_DESC) ? 
	      ainfo->sqinfo[idx].desc : "");
      P7PrintTrace(fp, tr[idx], hmm, dsq[idx]);
    }
  fputs("\n", fp);
}



/* Function: save_countvectors()
 * 
 * Purpose:  Save emission/transition count vectors to a file.
 *           Used for gathering the data on which to train a
 *           prior (e.g. mixture Dirichlet, etc.)
 *           
 *           The format of the file is one vector per line:
 *           M <f> <f> ...: 20 match emission counts in order AC..WY.
 *           I <f> <f> ...: 20 insert emission counts in order AC..WY.
 *           T <f> <f> ...: 7 transition counts in order TMM, TMI, TMD, 
 *                            TIM, TII, TDM, TDD. (see structs.h)
 *           
 * Args:     cfile  - counts file to make
 *           hmm    - counts-based HMM
 */
static void
save_countvectors(char *cfile, struct plan7_s *hmm)
{
  FILE *fp;
  int k, x;

  if ((fp = fopen(cfile, "w")) == NULL)
    Die("failed to open count vector file %s for writing", cfile);

				/* match emission vectors */
  for (k = 1; k <= hmm->M; k++)
    {
      fputs("M ", fp);
      for (x = 0; x < Alphabet_size; x++)
	fprintf(fp, "%.2f ", hmm->mat[k][x]);
      fputs("\n", fp);
    }
				/* insert emission vectors */
  for (k = 1; k < hmm->M; k++)
    {
      fputs("I ", fp);
      for (x = 0; x < Alphabet_size; x++)
	fprintf(fp, "%.2f ", hmm->ins[k][x]);
      fputs("\n", fp);
    }
				/* transition vectors */
    for (k = 1; k < hmm->M; k++)
    {
      fputs("T ", fp);
      for (x = 0; x < 7; x++)
	fprintf(fp, "%.2f ", hmm->t[k][x]);
      fputs("\n", fp);
    }

  fclose(fp);
}


/* Function: position_average_score()
 * Date:     Wed Dec 31 09:36:35 1997 [StL]
 * 
 * Purpose:  Calculate scores from tracebacks, keeping them
 *           in a position specific array. The final array
 *           is normalized position-specifically too, according
 *           to how many sequences contributed data to this
 *           position. Used for compensating for sequence 
 *           fragments in ME and MD score optimization. 
 *           Very much ad hoc.
 *           
 *           Code related to (derived from) TraceScore().
 *           
 * Args:     hmm       - HMM structure, scores valid
 *           dsq       - digitized unaligned sequences
 *           wgt       - weights on the sequences
 *           nseq      - number of sequences
 *           tr        - array of nseq tracebacks that aligns each dsq to hmm
 *           pernode   - RETURN: [0]1..M array of position-specific avg scores
 *           ret_avg   - RETURN: overall average full-length, one-domain score
 *           
 * Return:   1 on success, 0 on failure.          
 *           pernode is malloc'ed [0]1..M by CALLER and filled here.
 */
static void
position_average_score(struct plan7_s    *hmm, 
		       char             **dsq, 
		       float             *wgt,
		       int                nseq,
		       struct p7trace_s **tr,
		       float             *pernode,
		       float             *ret_avg)
{
  int    pos;                   /* position in seq */
  int    sym;
  int    tpos;                  /* position in trace/state sequence */
  float *counts;                /* counts at each position */
  float  avg;                   /* RETURN: average overall */
  int    k;                     /* counter for model position */
  int    idx;                   /* counter for sequence number */

  /* Allocations
   */
  counts = MallocOrDie ((hmm->M+1) * sizeof(float));
  FSet(pernode, hmm->M+1, 0.);
  FSet(counts,  hmm->M+1, 0.);

  /* Loop over traces, accumulate weighted scores per position
   */
  for (idx = 0; idx < nseq; idx++)
    for (tpos = 0; tpos < tr[idx]->tlen; tpos++)
      {
	pos = tr[idx]->pos[tpos];
	sym = (int) dsq[idx][tr[idx]->pos[tpos]];
	k   = tr[idx]->nodeidx[tpos];

	/* Counts: how many times did we use this model position 1..M?
         * (weighted)
	 */
	if (tr[idx]->statetype[tpos] == STM || tr[idx]->statetype[tpos] == STD)
	  counts[k] += wgt[idx];

	/* Emission scores.
	 */
	if (tr[idx]->statetype[tpos] == STM) 
	  pernode[k] += wgt[idx] * Scorify(hmm->msc[sym][k]);
	else if (tr[idx]->statetype[tpos] == STI) 
	  pernode[k] += wgt[idx] * Scorify(hmm->isc[sym][k]);
	
	/* Transition scores.
	 */
	if (tr[idx]->statetype[tpos] == STM ||
	    tr[idx]->statetype[tpos] == STD ||
	    tr[idx]->statetype[tpos] == STI)
	  pernode[k] += wgt[idx] * 
	    Scorify(TransitionScoreLookup(hmm, tr[idx]->statetype[tpos], tr[idx]->nodeidx[tpos],
					  tr[idx]->statetype[tpos+1],tr[idx]->nodeidx[tpos+1]));
      }

  /* Divide accumulated scores by accumulated weighted counts
   */
  avg = 0.;
  for (k = 1; k <= hmm->M; k++)
    {
      pernode[k] /= counts[k];
      avg += pernode[k];
    }
  
  free(counts);
  *ret_avg = avg;
  return;
}


/* Function: frag_trace_score()
 * Date:     SRE, Wed Dec 31 10:03:47 1997 [StL]
 * 
 * Purpose:  Allow MD/ME optimization to be used for alignments
 *           that include fragments and multihits -- estimate a full-length
 *           per-domain score.
 *           
 *
 *           
 * Return:   "corrected" score.
 */
static float
frag_trace_score(struct plan7_s *hmm, char *dsq, struct p7trace_s *tr, 
                 float *pernode, float expected)
{
  float sc;			/* corrected score  */
  float fragexp;		/* expected score for a trace like this */
  int   tpos;			/* position in trace */

                                /* get uncorrected score */
  sc = P7TraceScore(hmm, dsq, tr);

                               /* calc expected score for trace like this */
  fragexp = 0.;
  for (tpos = 0; tpos < tr->tlen; tpos++)
    if (tr->statetype[tpos] == STM || tr->statetype[tpos] == STD)
      fragexp += pernode[tr->nodeidx[tpos]];

				/* correct for multihits */
  fragexp /= (float) TraceDomainNumber(tr);

                                /* extrapolate to full-length, one-hit score */
  sc = sc * expected / fragexp;
  return sc;
}


/* Function: maximum_entropy()
 * Date:     SRE, Fri Jan  2 10:56:00 1998 [StL]
 * 
 * Purpose:  Optimizes a model according to maximum entropy weighting.
 *           See Krogh and Mitchison (1995).
 *
 *           [Actually, we do minimum relative entropy, rather than
 *           maximum entropy. Same thing, though we refer to "ME"
 *           weights and models. The optimization is a steepest
 *           descents minimization of the relative entropy.]
 *           
 *           Expects to be called shortly after a Maxmodelmaker()
 *           or Handmodelmaker(), so that both a new model architecture
 *           (with MAP parameters) and fake tracebacks are available.
 *           
 *           Prints a summary of optimization progress to stdout.
 *           
 * Args:     hmm     - model. allocated, set with initial MAP parameters.
 *           dsq     - dealigned digitized seqs the model is based on
 *           ainfo   - extra info for aseqs
 *           nseq    - number of aseqs
 *           eff_nseq- effective sequence number; weights normalize up to this.
 *           prior   - prior distributions for parameterizing model
 *           tr      - array of fake traces for each sequence        
 *           
 * Return:   (void)
 *           hmm changed to an ME HMM
 *           ainfo changed, contains ME weights          
 */
static void
maximum_entropy(struct plan7_s *hmm, char **dsq, AINFO *ainfo, int nseq,
		float eff_nseq, struct p7prior_s *prior, struct p7trace_s **tr)
{
  float *wgt;                  /* current best set of ME weights   */
  float *new_wgt;              /* new set of ME weights to try     */
  float *sc;                    /* log-odds score of each sequence */
  float *grad;                  /* gradient                        */
  float  epsilon;               /* steepness of descent            */
  float  relative_entropy;      /* current best relative entropy   */
  float  new_entropy;           /* relative entropy at new weights */
  float  last_new_entropy;      /* last new_entropy we calc'ed     */
  float  use_epsilon;           /* current epsilon value in use    */
  int    idx;                   /* counter over sequences          */
  int    i1, i2;		/* counters for iterations         */

  float  converge_criterion;
  float  minw, maxw;            /* min, max weight                 */
  int    posw, highw;           /* number of positive weights      */
  float  mins, maxs, avgs;      /* min, max, avg score             */
  float *pernode;               /* expected score per node of HMM  */
  float  expscore;              /* expected score of complete HMM  */
  int    max_iter;		/* bulletproof against infinite loop bugs */

  epsilon  = 0.2;                /* works fine */
  max_iter = 666;
  
  /* Allocations
   */
  sc      = MallocOrDie (sizeof(float) * nseq);
  wgt     = MallocOrDie (sizeof(float) * nseq);
  new_wgt = MallocOrDie (sizeof(float) * nseq);
  grad    = MallocOrDie (sizeof(float) * nseq);
  pernode = MallocOrDie (sizeof(float) * (hmm->M+1));

  /* Initialization. Start with all weights == 1.0.
   * Find relative entropy and gradient.
   */
  Plan7SWConfig(hmm, 0.5, 0.5);
  P7Logoddsify(hmm, TRUE);

  FSet(wgt, nseq, 1.0);
  position_average_score(hmm, dsq, wgt, nseq, tr, pernode, &expscore);
  for (idx = 0; idx < nseq; idx++) 
    sc[idx] = frag_trace_score(hmm, dsq[idx], tr[idx], pernode, expscore);
  relative_entropy = FSum(sc, nseq) / (float) nseq;
  for (idx = 0; idx < nseq; idx++)
    grad[idx] = relative_entropy - sc[idx];

  
  printf("iter avg-sc min-sc max-sc min-wgt max-wgt +wgt ++wgt rel.ent convergence\n");
  printf("---- ------ ------ ------ ------- ------- ---- ----- ------- -----------\n");
  mins = maxs = avgs = sc[0];
  for (idx = 1; idx < nseq; idx++)
    {
      if (sc[idx] < mins) mins = sc[idx];
      if (sc[idx] > maxs) maxs = sc[idx];
      avgs += sc[idx];
    }
  avgs /= nseq;
  printf("%4d %6.1f %6.1f %6.1f %7.2f %7.2f %4d %5d %7.2f %8s\n",
         0, avgs, mins, maxs, 1.0, 1.0, nseq, 0, relative_entropy, "-");

  
  /* Steepest descents optimization;
   * iterate until relative entropy converges.
   */
  i1 = 0;
  while (++i1 < max_iter)
    {
      /* Gradient gives us a line of steepest descents.
       * (Roughly speaking, anyway. We actually have a constraint
       * that weights are nonnegative and normalized, and the
       * gradient doesn't take these into account.)
       * Look along this line, a distance of epsilon * gradient:
       * if new point is better, accept; if new point is worse,
       * move back along the line by half the distance and re-evaluate.
       */
      use_epsilon = epsilon;
      new_entropy = relative_entropy + 1.0;    /* just ensure new > old */

      i2 = 0; 
      while (new_entropy > relative_entropy && ++i2 < max_iter)
        {
          last_new_entropy = new_entropy;

                                /* find a new point in weight space */
          for (idx = 0; idx < nseq; idx++)
            {
              new_wgt[idx] = wgt[idx] + use_epsilon * grad[idx];
              if (new_wgt[idx] < 0.) new_wgt[idx] = 0.0;
            }
          FNorm(new_wgt, nseq);
          FScale(new_wgt, nseq, (float) nseq);

                                /* Make new HMM using these weights */
          ZeroPlan7(hmm);
          for (idx = 0; idx < nseq; idx++)
            P7TraceCount(hmm, dsq[idx], new_wgt[idx], tr[idx]);
          P7PriorifyHMM(hmm, prior);

  
                                /* Evaluate new point */
	  Plan7SWConfig(hmm, 0.5, 0.5);
	  P7Logoddsify(hmm, TRUE);
	  position_average_score(hmm, dsq, new_wgt, nseq, tr, pernode, &expscore);
          for (idx = 0; idx < nseq; idx++) 
	    sc[idx]      = frag_trace_score(hmm, dsq[idx], tr[idx], pernode, expscore);
	  new_entropy = FDot(sc, new_wgt, nseq) / nseq;

          use_epsilon /= 2.0;
	  /* Failsafe: we're not converging. Set epsilon to zero,
	   * do one more round.
	   */
	  if (use_epsilon < 1e-6) use_epsilon = 0.0; 
	  if (use_epsilon == 0.0) break;
          
          /* Failsafe: avoid infinite loops. Sometimes the
             new entropy converges without ever being better 
             than the previous point, probably as a result
             of minor roundoff error. */
          if (last_new_entropy == new_entropy) break;
        }
      if (i2 == max_iter) printf("   -- exceeded maximum iterations; giving up --\n");

      /* Evaluate convergence before accepting the new weights;
       * then, accept the new point and evaluate the gradient there.
       */
      converge_criterion = fabs((relative_entropy-new_entropy)/relative_entropy);
      relative_entropy = new_entropy;
      FCopy(wgt, new_wgt, nseq);
      for (idx = 0; idx < nseq; idx++)
	grad[idx] = relative_entropy - sc[idx];

      /* Print some statistics about this iteration
       */
      mins = maxs = avgs = sc[0];
      minw = maxw = wgt[0];
      posw = (wgt[0] > 0.0) ? 1 : 0;
      highw = (wgt[0] > 1.0) ? 1 : 0;
      for (idx = 1; idx < nseq; idx++)
        {
          if (sc[idx] < mins) mins = sc[idx];
          if (sc[idx] > maxs) maxs = sc[idx];
          if (wgt[idx] < minw) minw = wgt[idx];
          if (wgt[idx] > maxw) maxw = wgt[idx];
          if (wgt[idx] > 0.0)  posw++;
          if (wgt[idx] > 1.0)  highw++;
          avgs += sc[idx];
        }
      avgs /= nseq;
      printf("%4d %6.1f %6.1f %6.1f %7.2f %7.2f %4d %5d %7.2f %8.5f\n",
             i1, 
             avgs, mins, maxs, 
             minw, maxw, posw, highw,
             relative_entropy, converge_criterion);
      
      if (converge_criterion < 1e-5) break;
    }
  if (i1 == max_iter) printf("   -- exceeded maximum iterations; giving up --\n");

  /* Renormalize weights to sum to eff_nseq, and save.
   */
  FNorm(wgt, nseq);
  FScale(wgt, nseq, (float) eff_nseq);
  FCopy(ainfo->wgt, wgt, nseq);
			/* Make final HMM using these adjusted weights */
  ZeroPlan7(hmm);
  for (idx = 0; idx < nseq; idx++)
    P7TraceCount(hmm, dsq[idx], wgt[idx], tr[idx]);
  P7PriorifyHMM(hmm, prior);
                                
  /* Cleanup and return
   */
  free(pernode);
  free(new_wgt);
  free(grad);
  free(wgt);
  free(sc);
  return;
}