bcload.c

BCView - Bayes Classifier Visualization Download xbcview Linux executable (218 kb) wbcview.exe W

    else if (attinfo.v_att < 0) /* note the second numeric attribute */
      attinfo.v_att = attcnt;   /* for the vertical direction */
    names [attcnt]   = att_name(att);
    attmap[attcnt++] = k;       /* note the attribute identifier */
    if (type == AT_INT) {       /* if attribute is integer valued */
      i = att_valmin(att)->i;   /* get and set minimum value */
      rng->min = (i <=  INT_MIN +1) ? -DBL_MAX : (double)i;
      i = att_valmax(att)->i;   /* get and set maximum value */
      rng->max = (i >=  INT_MAX)    ?  DBL_MAX : (double)i; }
    else {                      /* if attribute is real valued */
      f = att_valmin(att)->f;   /* get and set minimum value */
      rng->min = (f <= -FLT_MAX)    ? -DBL_MAX : (double)f;
      f = att_valmax(att)->f;   /* get and set maximum value */
      rng->max = (f >=  FLT_MAX)    ?  DBL_MAX : (double)f;
    }                           /* (set initial range of values ) */
    for (i = clscnt; --i >= 0;){/* traverse the normal distributions */
      if      (nbc) {           /* if naive Bayes classifier */
        exp =      nbc_exp(nbc, i, k);
        dev = sqrt(nbc_var(nbc, i, k)); }
      else if (fbc) {           /* if full Bayes classifier */
        mvn = fbc_mvnorm(fbc, i);
        exp =      mvn_exp(mvn, k);
        dev = sqrt(mvn_var(mvn, k)); }
      else {                    /* if cluster set */
        r   = am_off(cls_attmap(clset), k);
        exp = cls_center(clset, i)[r];
        mat = cls_covmat(clset, i);
        dev = sqrt((mat) ? mat_get(mat, r, r) : cls_isovar(clset, i));
      }                         /* get exp. value and std. deviation */
      t = exp -3 *dev; if (t < rng->min) rng->min = t;
      t = exp +3 *dev; if (t > rng->max) rng->max = t;
    }                           /* adapt the value range */
    if (rng->min >= rng->max) { rng->min = 0; rng->max = 1; }
    rng++;                      /* check and adapt final range */
  }                             /* and go to the next attribute */
  if (attinfo.v_att < 0) attinfo.v_att = attinfo.h_att;
  attinfo.h_rng = ranges[attinfo.h_att];
  attinfo.v_rng = ranges[attinfo.v_att];
  return 0;                     /* get ranges of selected attributes */
}  /* bc_load() */              /* and return `ok' */

/*--------------------------------------------------------------------*/

int bc_data (const char *fname)
{                               /* --- load data table */
  int    i, r, f;               /* loop variable, buffers */
  FILE   *file;                 /* input file */
  TFSCAN *tfs;                  /* table file scanner */
  ATT    *att;                  /* class attribute */

  if (!attset) return E_DATA;   /* check for an attribute set */
  for (i = as_attcnt(attset); --i >= 0; )
    att_setmark(as_att(attset, i), 1);
  if (nbc || fbc) {             /* mark all attributes and if class. */
    att = as_att(attset, (nbc) ? nbc_clsid(nbc) : fbc_clsid(fbc));
    att_setmark(att, 0);        /* get the class attribute, */
    att_inst(att)->i = UV_SYM;  /* mark it as optional, and */
  }                             /* set its value to unknown */
  /* With these preparations and the flag AS_MARKED in the call to */
  /* as_read the class is read only if it is present in the data.  */
  /* Otherwise the class is set to unknown in all tuples.          */

  file = fopen(fname, "r");     /* open table file for input */
  if (!file) return E_FOPEN;    /* and set the format characters */
  as_chars(attset, fmtinfo.blanks,  fmtinfo.fldseps,
                   fmtinfo.recseps, fmtinfo.uvchars);
  recno = 1;                    /* initialize the record number */
  if (fmtinfo.first == FR_COMMENT) {
    tfs = as_tfscan(attset);    /* if to skip a comment line, */
    do {                        /* loop to skip fields */
      r = tfs_getfld(tfs, file, buf, sizeof(buf)-1);
      if (r < 0) { fclose(file); return E_FREAD; }
    } while (r > TFS_REC);      /* ignore all fields up to */
    recno++;                    /* the end of the first record */
  }
  f = ((fmtinfo.first == FR_ATTS)  /* build read flags */
    ? AS_ATT : AS_DFLT)|AS_MARKED|AS_NOXATT;
  if (as_read(attset, file, f) != 0) {
    fclose(file);               /* read the first record */
    return as_err(attset)->code;/* of the data file */
  }                             /* (attribute names/first tuple) */
  if (table)                    /* if a table exists, delete it, */
    tab_delete(table, 0);       /* then create a new table */
  table = tab_create("table", attset, tpl_delete);
  if (!table) { fclose(file); return E_NOMEM; }
  r = (f & AS_ATT) ? as_read(attset, file, AS_INST) : 0;
  if (f & AS_ATT) recno++;      /* read the first data record */
  while (r == 0) {              /* record read loop */
    if (tab_tpladd(table, NULL) != 0) {
      r = E_NOMEM; break; }     /* add read tuple to the table */
    recno++;                    /* count the processed record */
    r = as_read(attset, file, AS_INST);
  }                             /* read the next tuple */
  fclose(file);                 /* close the input file and */
  if (r < 0) {                  /* check for a read error */
    tab_delete(table, 0); table = NULL; return r; }
  return 0;                     /* return `ok' */
}  /* bc_data() */

/*--------------------------------------------------------------------*/

int bc_bvnorm (void)
{                               /* --- init. normal distributions */
  int    i, k;                  /* loop variables */
  int    x, y;                  /* attribute identifiers */
  int    clscnt;                /* number of classes */
  MVNORM *mvn;                  /* to traverse the normal distribs. */
  double *ctr;                  /* to traverse the cluster centers */
  MATRIX *cov;                  /* to traverse the covariance mats. */
  double t0, t1, t2;            /* temporary buffers */

  x = attmap[attinfo.h_att];    /* get the attribute identifiers */
  y = attmap[attinfo.v_att];    /* of the selected attributes */
  if (nbc) {                    /* if naive Bayes classifier, */
    clscnt = nbc_clscnt(nbc);   /* get the number of classes */
    for (i = k = 0; i < clscnt; i++) {
      k |= bvn_init(bvnorm +i, nbc_prior(nbc, i),
                    nbc_exp(nbc, i, x), nbc_exp(nbc, i, y),
                    nbc_var(nbc, i, x), nbc_var(nbc, i, y), 0);
    } }                         /* create bivariate normal dists. */
  else if (fbc) {               /* if full Bayes classifier, */
    clscnt = fbc_clscnt(fbc);   /* get the number of classes */
    for (i = k = 0; i < clscnt; i++) {
      mvn = fbc_mvnorm(fbc, i); /* traverse the normal distributions */
      k |= bvn_init(bvnorm +i, fbc_prior(fbc, i),
                    mvn_exp(mvn, x), mvn_exp(mvn, y),
                    mvn_var(mvn, x), mvn_var(mvn, y),
                    (x != y) ? mvn_cov(mvn, x, y) : 0);
    } }                         /* create bivariate normal dists. */
  else {                        /* if cluster set */
    x = am_off(cls_attmap(clset), x);  /* get the rows/columns */
    y = am_off(cls_attmap(clset), y);  /* associated with the atts. */
    clscnt = cls_clscnt(clset); /* get the number of classes */
    for (i = k = 0; i < clscnt; i++) {
      ctr = cls_center(clset, i);  /* traverse the clusters */
      cov = cls_covmat(clset, i);  /* and get their parameters */
      if (!cov) {               /* if only isotropic variance */
        t0 = t1 = cls_isovar(clset, i); t2 = 0; }
      else {                    /* if direction dependent variance */
        t0 = mat_get(cov, x, x);
        t1 = mat_get(cov, y, y);
        t2 = (x != y) ? mat_get(cov, x, y) : 0;
      }                         /* get variances and covariance */
      k |= bvn_init(bvnorm +i, cls_weight(clset, i),
                    ctr[x], ctr[y], t0, t1, t2);
    }                           /* create a bivariate normal dist. */
  }                             /* for each of the clusters */
  return (k != 0) ? -1 : clscnt;/* return the initialization status */
}  /* bc_bvnorm() */