/* Last edited on 2010-05-25 04:48:49 by stolfi  */

pnm_image_t *rn_classif_pnm_image_from_dataset
  ( int NA,
    int NCL, 
    rn_classif_labeler_t *lab, 
    r2_t *ictr, 
    double HV, 
    int NXY, 
    int NSUB, 
    double sigma
  );
  /* Produces an image with a pixel-size dot for each element of the dataset
    {D}. The dots are colored according to their {classD} values, in the same
    color scheme.  Assumes {classD} ranges over {0..NCD}.
    Requires {NA == 2} and {D.NA >= 2}, and plots only the first 2 attributes of each sample
    against a white background.
    
    If {lab2} is not NULL, the brightness of the background
    is modified according to the class {cl2} assigned by
    {(NA,NCL,lab)}. If {cl2 = 0} it remains white, otherwise it is 
    replced by darker shades of gray.
    
    The parameters {ictr,HV,NXY,NSUB,sigma} have the same meaning 
    as for {rn_classif_pnm_image_from_labeler}, except that {sigma}
    and {NSUB} apply only to the background painting with {NA,NCL,lab}.
    The samples of {D} are not perturbed by the {sigma}-noise
    and their dots are anti-aliased in any case. */
                  
pnm_image_t *rn_classif_pnm_image_from_dataset
  ( int NA,
    int NCD, 
    rn_classif_dataset_t *D, 
    int classD[],
    int NCL, 
    rn_classif_labeler_t *lab, 
    r2_t *ictr, 
    double HV, 
    int NXY, 
    int NSUB, 
    double sigma
  )
  { 
    demand(NA == 2, "problem is not two-dimensional");
    demand(D->NA >= NA, "not enough attributes in dataset");
    
    /* Assign colors to classes: */
    frgb_t *cmap = rn_classif_pick_class_colors(NCD);
    
    float_image_t *fim = float_image_new(3, NXY, NXY);
    
    /* Paint a white background shadowed by {lab}: */
    rn_classif_float_image_paint_labeler(fim, NA, 0, NULL, NCL, lab, ictr, HV, NSUB, sigma, cmap);
      
    /* Paint the dots over it: */
    rn_classif_float_image_paint_dataset(fim, NA, NCD, D, classD, ictr, HV, 3, cmap);
  
    pnm_image_t *pim = float_image_to_pnm_image(fim, 3, NULL, NULL, NULL, 255, TRUE, FALSE);
    float_image_free(fim);
    free(cmap);
    return pim;
  }

auto void debug_frgb(frgb_t val, char *spc);
    void debug_frgb(frgb_t val, char *spc)
      { if (spc == NULL) 
          { fprintf(stderr, "\n"); }
        else
          { fprintf(stderr, "  %s ( %6.4f %6.4f %6.4f )\n", spc, val.c[0], val.c[1], val.c[2]); }
      }
    
    /* bool_t debug = (r2_norm(pctr) <= 1.1) && (lab2 != NULL); */
            bool_t debug = FALSE;
            if (debug) { debug_frgb(val, "RGB"); }
            if (debug) { debug_frgb(val, "HTY"); }
                if (debug) { debug_frgb(val, "HTY"); }
                if (debug) { debug_frgb(val, "RGB"); }
                if (debug) { debug_frgb(val, NULL); }
                

typedef struct rn_classif_sample_t 
  { int class;       /* The nominal class of the sample, or 0 for `no clss'. */
    float weight;    /* A parameter for the distance function. */
    double *attr;    /* {attr[t]} is attribute {t} of the sample. */
  } rn_classif_sample_t;
  /* A sample vector with a weight and a nominal classification. */