STATS Class Reference

#include <statistc.h>

Public Member Functions
	STATS (inT32 min_bucket_value, inT32 max_bucket_value_plus_1)
	STATS ()
	~STATS ()
bool	set_range (inT32 min_bucket_value, inT32 max_bucket_value_plus_1)
void	clear ()
void	add (inT32 value, inT32 count)
inT32	mode () const
double	mean () const
double	sd () const
double	ile (double frac) const
inT32	min_bucket () const
inT32	max_bucket () const
double	median () const
inT32	pile_count (inT32 value) const
inT32	get_total () const
bool	local_min (inT32 x) const
void	smooth (inT32 factor)
inT32	cluster (float lower, float upper, float multiple, inT32 max_clusters, STATS *clusters)
int	top_n_modes (int max_modes, GenericVector< tesseract::KDPairInc< float, int > > *modes) const
void	print () const
void	print_summary () const
void	plot (ScrollView *window, float xorigin, float yorigin, float xscale, float yscale, ScrollView::Color colour) const
void	plotline (ScrollView *window, float xorigin, float yorigin, float xscale, float yscale, ScrollView::Color colour) const

Detailed Description

Definition at line 33 of file statistc.h.

Constructor & Destructor Documentation

STATS() [1/2]

STATS::STATS	(	inT32	min_bucket_value,
		inT32	max_bucket_value_plus_1
	)

Definition at line 40 of file statistc.cpp.

                                                                  {
  if (max_bucket_value_plus_1 <= min_bucket_value) {
    min_bucket_value = 0;
    max_bucket_value_plus_1 = 1;
  }
  rangemin_ = min_bucket_value;                // setup
  rangemax_ = max_bucket_value_plus_1;
  buckets_ = new inT32[rangemax_ - rangemin_];
  clear();
}

STATS() [2/2]

STATS::STATS

(

)

Definition at line 51 of file statistc.cpp.

             {
  rangemax_ = 0;
  rangemin_ = 0;
  buckets_ = NULL;
}

~STATS()

STATS::~STATS

(

)

Definition at line 92 of file statistc.cpp.

               {
  delete [] buckets_;
}

Member Function Documentation

add()

void STATS::add	(	inT32	value,
		inT32	count
	)

Definition at line 101 of file statistc.cpp.

                                        {
  if (buckets_ == NULL) {
    return;
  }
  value = ClipToRange(value, rangemin_, rangemax_ - 1);
  buckets_[value - rangemin_] += count;
  total_count_ += count;          // keep count of total
}

clear()

void STATS::clear

(

)

Definition at line 81 of file statistc.cpp.

                  {  // clear out buckets
  total_count_ = 0;
  if (buckets_ != NULL)
    memset(buckets_, 0, (rangemax_ - rangemin_) * sizeof(buckets_[0]));
}

cluster()

inT32 STATS::cluster	(	float	lower,
		float	upper,
		float	multiple,
		inT32	max_clusters,
		STATS *	clusters
	)

Definition at line 320 of file statistc.cpp.

                                      {   // array of clusters
  BOOL8 new_cluster;             // added one
  float *centres;                // cluster centres
  inT32 entry;                   // bucket index
  inT32 cluster;                 // cluster index
  inT32 best_cluster;            // one to assign to
  inT32 new_centre = 0;          // residual mode
  inT32 new_mode;                // pile count of new_centre
  inT32 count;                   // pile to place
  float dist;                    // from cluster
  float min_dist;                // from best_cluster
  inT32 cluster_count;           // no of clusters

  if (buckets_ == NULL || max_clusters < 1)
    return 0;
  centres = new float[max_clusters + 1];
  for (cluster_count = 1; cluster_count <= max_clusters
       && clusters[cluster_count].buckets_ != NULL
       && clusters[cluster_count].total_count_ > 0;
       cluster_count++) {
    centres[cluster_count] =
      static_cast<float>(clusters[cluster_count].ile(0.5));
    new_centre = clusters[cluster_count].mode();
    for (entry = new_centre - 1; centres[cluster_count] - entry < lower
         && entry >= rangemin_
         && pile_count(entry) <= pile_count(entry + 1);
         entry--) {
      count = pile_count(entry) - clusters[0].pile_count(entry);
      if (count > 0) {
        clusters[cluster_count].add(entry, count);
        clusters[0].add (entry, count);
      }
    }
    for (entry = new_centre + 1; entry - centres[cluster_count] < lower
         && entry < rangemax_
         && pile_count(entry) <= pile_count(entry - 1);
         entry++) {
      count = pile_count(entry) - clusters[0].pile_count(entry);
      if (count > 0) {
        clusters[cluster_count].add(entry, count);
        clusters[0].add(entry, count);
      }
    }
  }
  cluster_count--;

  if (cluster_count == 0) {
    clusters[0].set_range(rangemin_, rangemax_);
  }
  do {
    new_cluster = FALSE;
    new_mode = 0;
    for (entry = 0; entry < rangemax_ - rangemin_; entry++) {
      count = buckets_[entry] - clusters[0].buckets_[entry];
      //remaining pile
      if (count > 0) {           //any to handle
        min_dist = static_cast<float>(MAX_INT32);
        best_cluster = 0;
        for (cluster = 1; cluster <= cluster_count; cluster++) {
          dist = entry + rangemin_ - centres[cluster];
          //find distance
          if (dist < 0)
            dist = -dist;
          if (dist < min_dist) {
            min_dist = dist;     //find least
            best_cluster = cluster;
          }
        }
        if (min_dist > upper     //far enough for new
          && (best_cluster == 0
          || entry + rangemin_ > centres[best_cluster] * multiple
        || entry + rangemin_ < centres[best_cluster] / multiple)) {
          if (count > new_mode) {
            new_mode = count;
            new_centre = entry + rangemin_;
          }
        }
      }
    }
                                 // need new and room
    if (new_mode > 0 && cluster_count < max_clusters) {
      cluster_count++;
      new_cluster = TRUE;
      if (!clusters[cluster_count].set_range(rangemin_, rangemax_)) {
        delete [] centres;
        return 0;
      }
      centres[cluster_count] = static_cast<float>(new_centre);
      clusters[cluster_count].add(new_centre, new_mode);
      clusters[0].add(new_centre, new_mode);
      for (entry = new_centre - 1; centres[cluster_count] - entry < lower
        && entry >= rangemin_
      && pile_count (entry) <= pile_count(entry + 1); entry--) {
        count = pile_count(entry) - clusters[0].pile_count(entry);
        if (count > 0) {
          clusters[cluster_count].add(entry, count);
          clusters[0].add(entry, count);
        }
      }
      for (entry = new_centre + 1; entry - centres[cluster_count] < lower
        && entry < rangemax_
      && pile_count (entry) <= pile_count(entry - 1); entry++) {
        count = pile_count(entry) - clusters[0].pile_count(entry);
        if (count > 0) {
          clusters[cluster_count].add(entry, count);
          clusters[0].add (entry, count);
        }
      }
      centres[cluster_count] =
        static_cast<float>(clusters[cluster_count].ile(0.5));
    }
  } while (new_cluster && cluster_count < max_clusters);
  delete [] centres;
  return cluster_count;
}

get_total()

inT32 STATS::get_total ( ) const

inline

Definition at line 86 of file statistc.h.

                          {
    return total_count_;        // total of all piles
  }

ile()

double STATS::ile

(

double

frac

)

const

Definition at line 174 of file statistc.cpp.

                                   {
  if (buckets_ == NULL || total_count_ == 0) {
    return static_cast<double>(rangemin_);
  }
#if 0
  // TODO(rays) The existing code doesn't seem to be doing the right thing
  // with target a double but this substitute crashes the code that uses it.
  // Investigate and fix properly.
  int target = IntCastRounded(frac * total_count_);
  target = ClipToRange(target, 1, total_count_);
#else
  double target = frac * total_count_;
  target = ClipToRange(target, 1.0, static_cast<double>(total_count_));
#endif
  int sum = 0;
  int index = 0;
  for (index = 0; index < rangemax_ - rangemin_ && sum < target;
       sum += buckets_[index++]);
  if (index > 0) {
    ASSERT_HOST(buckets_[index - 1] > 0);
    return rangemin_ + index -
        static_cast<double>(sum - target) / buckets_[index - 1];
  } else {
    return static_cast<double>(rangemin_);
  }
}

local_min()

bool STATS::local_min

(

inT32

x

)

const

Definition at line 262 of file statistc.cpp.

                                   {
  if (buckets_ == NULL) {
    return false;
  }
  x = ClipToRange(x, rangemin_, rangemax_ - 1) - rangemin_;
  if (buckets_[x] == 0)
    return true;
  inT32 index;                   // table index
  for (index = x - 1; index >= 0 && buckets_[index] == buckets_[x]; --index);
  if (index >= 0 && buckets_[index] < buckets_[x])
    return false;
  for (index = x + 1; index < rangemax_ - rangemin_ &&
       buckets_[index] == buckets_[x]; ++index);
  if (index < rangemax_ - rangemin_ && buckets_[index] < buckets_[x])
    return false;
  else
    return true;
}

max_bucket()

inT32 STATS::max_bucket

(

)

const

Definition at line 221 of file statistc.cpp.

                              {  // Find max
  if (buckets_ == NULL || total_count_ == 0) {
    return rangemin_;
  }
  inT32 max;
  for (max = rangemax_ - rangemin_ - 1; max > 0 && buckets_[max] == 0; max--);
  return rangemin_ + max;
}

mean()

double STATS::mean

(

)

const

Definition at line 135 of file statistc.cpp.

                         {  //get mean of samples
  if (buckets_ == NULL || total_count_ <= 0) {
    return static_cast<double>(rangemin_);
  }
  inT64 sum = 0;
  for (int index = rangemax_ - rangemin_ - 1; index >= 0; --index) {
    sum += static_cast<inT64>(index) * buckets_[index];
  }
  return static_cast<double>(sum) / total_count_ + rangemin_;
}

median()

double STATS::median

(

)

const

Definition at line 239 of file statistc.cpp.

                           {  //get median
  if (buckets_ == NULL) {
    return static_cast<double>(rangemin_);
  }
  double median = ile(0.5);
  int median_pile = static_cast<int>(floor(median));
  if ((total_count_ > 1) && (pile_count(median_pile) == 0)) {
    inT32 min_pile;
    inT32 max_pile;
    /* Find preceding non zero pile */
    for (min_pile = median_pile; pile_count(min_pile) == 0; min_pile--);
    /* Find following non zero pile */
    for (max_pile = median_pile; pile_count(max_pile) == 0; max_pile++);
    median = (min_pile + max_pile) / 2.0;
  }
  return median;
}

min_bucket()

inT32 STATS::min_bucket

(

)

const

Definition at line 206 of file statistc.cpp.

                              {  // Find min
  if (buckets_ == NULL || total_count_ == 0) {
    return rangemin_;
  }
  inT32 min = 0;
  for (min = 0; (min < rangemax_ - rangemin_) && (buckets_[min] == 0); min++);
  return rangemin_ + min;
}

mode()

inT32 STATS::mode

(

)

const

Definition at line 115 of file statistc.cpp.

                        {  // get mode of samples
  if (buckets_ == NULL) {
    return rangemin_;
  }
  inT32 max = buckets_[0];           // max cell count
  inT32 maxindex = 0;                // index of max
  for (int index = rangemax_ - rangemin_ - 1; index > 0; --index) {
    if (buckets_[index] > max) {
      max = buckets_[index];      // find biggest
      maxindex = index;
    }
  }
  return maxindex + rangemin_;    // index of biggest
}

pile_count()

inT32 STATS::pile_count ( inT32  value ) const

inline

Definition at line 78 of file statistc.h.

                                       {
    if (value <= rangemin_)
      return buckets_[0];
    if (value >= rangemax_ - 1)
      return buckets_[rangemax_ - rangemin_ - 1];
    return buckets_[value - rangemin_];
  }

plot()

void STATS::plot	(	ScrollView *	window,
		float	xorigin,
		float	yorigin,
		float	xscale,
		float	yscale,
		ScrollView::Color	colour
	)		const

Definition at line 585 of file statistc.cpp.

                                               {   // colour to draw in
  if (buckets_ == NULL) {
    return;
  }
  window->Pen(colour);

  for (int index = 0; index < rangemax_ - rangemin_; index++) {
    window->Rectangle( xorigin + xscale * index, yorigin,
      xorigin + xscale * (index + 1),
      yorigin + yscale * buckets_[index]);
  }
}

plotline()

void STATS::plotline	(	ScrollView *	window,
		float	xorigin,
		float	yorigin,
		float	xscale,
		float	yscale,
		ScrollView::Color	colour
	)		const

Definition at line 612 of file statistc.cpp.

                                                   {  // colour to draw in
  if (buckets_ == NULL) {
    return;
  }
  window->Pen(colour);
  window->SetCursor(xorigin, yorigin + yscale * buckets_[0]);
  for (int index = 0; index < rangemax_ - rangemin_; index++) {
    window->DrawTo(xorigin + xscale * index,
                   yorigin + yscale * buckets_[index]);
  }
}

print()

void STATS::print

(

)

const

Definition at line 534 of file statistc.cpp.

                        {
  if (buckets_ == NULL) {
    return;
  }
  inT32 min = min_bucket() - rangemin_;
  inT32 max = max_bucket() - rangemin_;

  int num_printed = 0;
  for (int index = min; index <= max; index++) {
    if (buckets_[index] != 0) {
      tprintf("%4d:%-3d ", rangemin_ + index, buckets_[index]);
      if (++num_printed % 8 == 0)
        tprintf ("\n");
    }
  }
  tprintf ("\n");
  print_summary();
}

print_summary()

void STATS::print_summary

(

)

const

Definition at line 560 of file statistc.cpp.

                                {
  if (buckets_ == NULL) {
    return;
  }
  inT32 min = min_bucket();
  inT32 max = max_bucket();
  tprintf("Total count=%d\n", total_count_);
  tprintf("Min=%.2f Really=%d\n", ile(0.0), min);
  tprintf("Lower quartile=%.2f\n", ile(0.25));
  tprintf("Median=%.2f, ile(0.5)=%.2f\n", median(), ile(0.5));
  tprintf("Upper quartile=%.2f\n", ile(0.75));
  tprintf("Max=%.2f Really=%d\n", ile(1.0), max);
  tprintf("Range=%d\n", max + 1 - min);
  tprintf("Mean= %.2f\n", mean());
  tprintf("SD= %.2f\n", sd());
}

sd()

double STATS::sd

(

)

const

Definition at line 151 of file statistc.cpp.

                       {  //standard deviation
  if (buckets_ == NULL || total_count_ <= 0) {
    return 0.0;
  }
  inT64 sum = 0;
  double sqsum = 0.0;
  for (int index = rangemax_ - rangemin_ - 1; index >= 0; --index) {
    sum += static_cast<inT64>(index) * buckets_[index];
    sqsum += static_cast<double>(index) * index * buckets_[index];
  }
  double variance = static_cast<double>(sum) / total_count_;
  variance = sqsum / total_count_ - variance * variance;
  if (variance > 0.0)
    return sqrt(variance);
  return 0.0;
}

set_range()

bool STATS::set_range	(	inT32	min_bucket_value,
		inT32	max_bucket_value_plus_1
	)

Definition at line 62 of file statistc.cpp.

                                                                           {
  if (max_bucket_value_plus_1 <= min_bucket_value) {
    return false;
  }
  if (rangemax_ - rangemin_ != max_bucket_value_plus_1 - min_bucket_value) {
    delete [] buckets_;
    buckets_ = new inT32[max_bucket_value_plus_1 - min_bucket_value];
  }
  rangemin_ = min_bucket_value;                // setup
  rangemax_ = max_bucket_value_plus_1;
  clear();                // zero it
  return true;
}

smooth()

void STATS::smooth

(

inT32

factor

)

Definition at line 289 of file statistc.cpp.

                               {
  if (buckets_ == NULL || factor < 2) {
    return;
  }
  STATS result(rangemin_, rangemax_);
  int entrycount = rangemax_ - rangemin_;
  for (int entry = 0; entry < entrycount; entry++) {
                                 //centre weight
    int count = buckets_[entry] * factor;
    for (int offset = 1; offset < factor; offset++) {
      if (entry - offset >= 0)
        count += buckets_[entry - offset] * (factor - offset);
      if (entry + offset < entrycount)
        count += buckets_[entry + offset] * (factor - offset);
    }
    result.add(entry + rangemin_, count);
  }
  total_count_ = result.total_count_;
  memcpy(buckets_, result.buckets_, entrycount * sizeof(buckets_[0]));
}

top_n_modes()

int STATS::top_n_modes	(	int	max_modes,
		GenericVector< tesseract::KDPairInc< float, int > > *	modes
	)		const

Definition at line 469 of file statistc.cpp.

                                                                           {
  if (max_modes <= 0) return 0;
  int src_count = rangemax_ - rangemin_;
  // Used copies the counts in buckets_ as they get used.
  STATS used(rangemin_, rangemax_);
  modes->truncate(0);
  // Total count of the smallest peak found so far.
  int least_count = 1;
  // Mode that is used as a seed for each peak
  int max_count = 0;
  do {
    // Find an unused mode.
    max_count = 0;
    int max_index = 0;
    for (int src_index = 0; src_index < src_count; src_index++) {
      int pile_count = buckets_[src_index] - used.buckets_[src_index];
      if (pile_count > max_count) {
        max_count = pile_count;
        max_index = src_index;
      }
    }
    if (max_count > 0) {
      // Copy the bucket count to used so it doesn't get found again.
      used.buckets_[max_index] = max_count;
      // Get the entire peak.
      double total_value = max_index * max_count;
      int total_count = max_count;
      int prev_pile = max_count;
      for (int offset = 1; max_index + offset < src_count; ++offset) {
        if (!GatherPeak(max_index + offset, buckets_, used.buckets_,
                        &prev_pile, &total_count, &total_value))
          break;
      }
      prev_pile = buckets_[max_index];
      for (int offset = 1; max_index - offset >= 0; ++offset) {
        if (!GatherPeak(max_index - offset, buckets_, used.buckets_,
                        &prev_pile, &total_count, &total_value))
          break;
      }
      if (total_count > least_count || modes->size() < max_modes) {
        // We definitely want this mode, so if we have enough discard the least.
        if (modes->size() == max_modes)
          modes->truncate(max_modes - 1);
        int target_index = 0;
        // Linear search for the target insertion point.
        while (target_index < modes->size() &&
               (*modes)[target_index].data >= total_count)
          ++target_index;
        float peak_mean =
            static_cast<float>(total_value / total_count + rangemin_);
        modes->insert(KDPairInc<float, int>(peak_mean, total_count),
                      target_index);
        least_count = modes->back().data;
      }
    }
  } while (max_count > 0);
  return modes->size();
}

---

The documentation for this class was generated from the following files:

• /home/stefan/src/github/tesseract-ocr/tesseract/ccstruct/statistc.h
• /home/stefan/src/github/tesseract-ocr/tesseract/ccstruct/statistc.cpp