*****************************************************************************/
#include "CSA.h"
#include "TextCollection.h"
+
+// Conflicts with std::min and std::max
+#undef min
+#undef max
+#include "dcover/bwt.hpp"
+
+#include "HeapProfiler.h" // FIXME remove
+
#include <iostream>
#include <queue>
-#include <iomanip>
#include <set>
#include <vector>
#include <utility>
using std::make_pair;
using std::map;
-using SXSI::TextCollection;
+namespace SXSI
+{
+// Save file version info
+const uchar CSA::versionFlag = 2;
////////////////////////////////////////////////////////////////////////////
// Class CSA::THuffAlphabetRank
-
+// FIXME Unused code
CSA::THuffAlphabetRank::THuffAlphabetRank(uchar *s, TextPosition n, TCodeEntry *codetable, unsigned level) {
left = NULL;
right = NULL;
}
-/////////////////////////////////////////////f///////////////////////////////
+////////////////////////////////////////////////////////////////////////////
// Class CSA
/**
* Samplerate defaults to TEXTCOLLECTION_DEFAULT_SAMPLERATE.
*/
CSA::CSA(unsigned samplerate)
- : n(0), alphabetrank(0), sampled(0), suffixes(0), positions(0),
- codetable(0), dynFMI(0), numberOfTexts(0), maxTextLength(0), endmarkerDocId(0),
- textLength(0), textStartPos(0)
+ : n(0), samplerate(0), alphabetrank(0), codetable(0), sampled(0), suffixes(0),
+ suffixDocId(0), numberOfTexts(0), numberOfAllTexts(0), maxTextLength(0), endmarkerDocId(0),
+ emptyTextRank(0)
{
this->samplerate = samplerate;
-
+
+#ifdef CSA_TEST_BWT
// FIXME TODO : DynFMI needs distribution of characters before hand
// This will create fully balanced wavelet tree for all chars [0, 255].
uchar temp[256];
uchar *temp = Tools::GetFileContents("data.txt", 0);
dynFMI = new DynFMI(temp,1790000, strlen((char *)temp), false);
delete [] temp;*/
+#endif
}
/**
void CSA::InsertText(uchar const * text)
{
// Sanity check:
+#ifdef CSA_TEST_BWT
assert(dynFMI != 0);
+#endif
TextPosition m = std::strlen((char *)text) + 1;
if (m > maxTextLength)
maxTextLength = m; // Store length of the longest text seen so far.
- this->n += m;
- this->numberOfTexts ++;
- dynFMI->addText(text, m);
+ if (m > 1)
+ {
+ this->n += m;
+ this->numberOfTexts ++;
+ this->numberOfAllTexts ++;
+#ifdef CSA_TEST_BWT
+ dynFMI->addText(text, m);
+#endif
+
+ texts.append((char *) text);
+ texts.append(1, '\0');
+ }
+ else
+ {
+ emptyTextId.insert(numberOfAllTexts); // FIXME Using too much space here
+ this->numberOfAllTexts ++;
+ }
}
void CSA::MakeStatic()
{
// Sanity check:
+#ifdef CSA_TEST_BWT
if (dynFMI == 0)
+ throw std::runtime_error("CSA::MakeStatic(): Data structure is already static (dynFMI == 0).");
+#endif
+
+ // Bitvector of empty texts, FIXME Remove?
{
- std::cerr << "Data structure is already static (dynFMI == 0)." << std::endl;
- std::exit(0);
+ //std::cout << std::endl << "texts: " << numberOfTexts << ", all texts " << numberOfAllTexts << ", size : " << emptyTextId.size() <<std::endl;
+ ulong *tempB = new ulong[numberOfAllTexts/W + 1];
+ for (ulong i = 0; i < numberOfAllTexts/W + 1; ++i)
+ tempB[i] = 0;
+ for (std::set<unsigned>::const_iterator it = emptyTextId.begin(); it != emptyTextId.end(); ++it)
+ Tools::SetField(tempB, 1, (*it), 1);
+ emptyTextId.clear();
+ emptyTextRank = new BSGAP(tempB, numberOfAllTexts, true);
}
-
- uchar *bwt = dynFMI->getBWT();
- /*printf("123456789012345678901234567890123456789\n");
- for (TextPosition i = 0; i < n; i ++)
- if (bwt[i] < 50)
- printf("%d", (int)bwt[i]);
- else
- printf("%c", bwt[i]);
- printf("\n");*/
-
-
-/* for (TextPosition i = 1; i <= n; i ++)
+
+ uchar *bwt = new uchar[n];
{
- printf("LF[%lu, %c] = %lu\n", i, (*dynFMI)[i], dynFMI->LFmapping(i));
- }*/
+ // More succinct solution via StringIterator, see below.
+/* unsigned* maptexts = new unsigned[n+1];
+ // Map text chars to [0..255+numberOfTexts]
+ unsigned count = 0;
+ for (ulong i = 0; i < n; ++i)
+ if (texts[i] == 0)
+ maptexts[i] = ++count; // endmarkers \in [1..numberOfTexts]
+ else {
+ uchar c = (uchar)texts[i];
+ maptexts[i] = (unsigned)c + numberOfTexts;
+ }
+ maptexts[n] = '\0';
+ assert(count == numberOfTexts);
+
+ std::cout << "maptext: ";
+ for (ulong i = 0; i <= n; ++i)
+ std::cout << maptexts[i] << ", ";
+ std::cout << std::endl;*/
+
+ // Mark endmarker positions into bitvector
+ ulong * endmarkers = new ulong[n/W+1];
+ for (ulong i = 0; i < n/W+1; ++i)
+ endmarkers[i] = 0;
+ for (ulong i = 0; i < n; ++i)
+ if (texts[i] == 0)
+ Tools::SetField(endmarkers, 1, i, 1);
+ BitRank* br = new BitRank(endmarkers, n, true);
+ assert(numberOfTexts == br->rank(n-1));
+
+ // Build iterators, FIXME clean up iterator construction.
+ StringIterator itBegin((uchar const *)texts.c_str(), (uchar const *)texts.c_str(), n, numberOfTexts, br);
+ StringIterator itEnd((uchar const *)texts.c_str() + n,(uchar const *)texts.c_str(), n, numberOfTexts, br);
- // Sanity check
- assert(numberOfTexts == dynFMI->getCollectionSize());
-
- delete dynFMI;
- dynFMI = 0;
+ bwtEndPos = (ulong)compute_bwt(itBegin, itEnd, //&maptexts[0], &maptexts[n],
+ &bwt[0], numberOfTexts);
- makewavelet(bwt);
+ bwt[--bwtEndPos] = '\0';
+ texts.erase();
+ texts.reserve(0); // Release capacity
+ delete br; // bitvector of endmarkers
+ } // End of bw transform
- // Calculate BWT end-marker position (of last inserted text)
- // and the length of the first text (counter l):
- ulong i = 0;
- ulong l = 1;
- while (bwt[i] != '\0')
- {
- uchar c = bwt[i];
- i = C[c]+alphabetrank->rank(c, i)-1;
- ++l;
+#ifdef CSA_TEST_BWT
+ {
+ uchar *bwtTest = dynFMI->getBWT();
+ printf("123456789012345678901234567890123456789\n");
+ for (TextPosition i = 0; i < n && i < 100; i ++)
+ if (bwt[i] < 50)
+ printf("%d", (int)bwt[i]);
+ else
+ printf("%c", bwt[i]);
+ printf("\n");
+ for (TextPosition i = 0; i < n && i < 100; i ++)
+ if (bwtTest[i] < 50)
+ printf("%d", (int)bwtTest[i]);
+ else
+ printf("%c", bwtTest[i]);
+ printf("\n");
+
+ // Sanity check
+ assert(numberOfTexts == dynFMI->getCollectionSize());
+
+ delete dynFMI;
+ dynFMI = 0;
+ for (ulong i = 0; i < n; ++i)
+ if (bwt[i] != bwtTest[i])
+ {
+ std::cout << "i = " << i << ", bwt = " << (unsigned)bwt[i] << ", " << (unsigned)bwtTest[i] << std::endl;
+ assert(0);
+ }
+ delete [] bwtTest;
}
- bwtEndPos = i;
- //printf("bwtEndPos = %lu\n", bwtEndPos);
-
- delete [] bwt;
+#endif // CSA_TEST_BWT
- // Build up arrays for text length and starting positions
- textLength = new BlockArray(numberOfTexts, Tools::CeilLog2(maxTextLength));
- textStartPos = new BlockArray(numberOfTexts, Tools::CeilLog2(this->n));
- (*textLength)[0] = l;
- (*textStartPos)[0] = 0; // Rest of the values are updated in CSA::maketables().
+ makewavelet(bwt); // Deletes bwt!
+ bwt = 0;
// Make sampling tables
maketables();
bool CSA::EmptyText(DocId k) const
{
- assert(k < (DocId)numberOfTexts);
- return (1 == (*textLength)[k]);
+ assert(k < (DocId)numberOfTexts);
+ if (emptyTextRank->IsBitSet(k))
+ return true;
+ return false;
}
uchar* CSA::GetText(DocId k) const
{
assert(k < (DocId)numberOfTexts);
- uchar* result = new uchar[(*textLength)[k]];
+ ulong textRank = 0;
+ if (emptyTextRank->IsBitSet(k, &textRank))
+ {
+ uchar* result = new uchar[1];
+ result[0] = 0;
+ return result;
+ }
+ // Map to non-empty text
+ k -= textRank; //emptyTextRank->rank(k);
+
TextPosition i = k;
- TextPosition j = (*textLength)[k] - 1;
- result[j] = '\0';
- uchar c = alphabetrank->charAtPos(i);
+ string result;
+ // Reserve average string length to avoid reallocs
+ result.reserve(n/numberOfTexts);
+
+ uchar c = alphabetrank->access(i);
while (c != '\0')
{
- --j;
- result[j] = c;
+ result.push_back(c);
i = C[c]+alphabetrank->rank(c,i)-1;
- c = alphabetrank->charAtPos(i); // "next" char.
+ c = alphabetrank->access(i); // "next" char.
}
- assert(j == 0);
- return result;
+
+ // Convert to uchar (FIXME return string?)
+ i = result.size();
+ uchar* res = new uchar[i+1];
+ res[i] = '\0';
+ for (ulong j = 0; j < i; ++j)
+ res[i-j-1] = result[j];
+ return res;
}
-
+/*
+ * Not supported
uchar* CSA::GetText(DocId k, TextPosition i, TextPosition j) const
{
assert(k < (DocId)numberOfTexts);
assert(j < (*textLength)[k]);
assert(i <= j);
+ ulong textRank = 0;
+ if (emptyTextRank->IsBitSet(k, &textRank))
+ {
+ uchar* result = new uchar[1];
+ result[0] = 0;
+ return result;
+ }
+
+ // Map to non-empty text
+ k -= textRank; // emptyTextRank->rank(k);
+
// Start position of k'th text
ulong start = (*textStartPos)[k];
return Substring(i + start, j-i+1);
-}
+ }*/
/******************************************************************
* Existential queries
bool CSA::IsSuffix(uchar const *pattern) const
{
- // Here counting is as fast as isSuffix():
+ // Here counting is as fast as IsSuffix():
if (CountSuffix(pattern) > 0)
return true;
return false;
bool CSA::IsEqual(uchar const *pattern) const
{
TextPosition m = std::strlen((char *)pattern);
- //if (m == 0)
- // return false; // FIXME Check for empty texts!
+ if (m == 0)
+ {
+ if (numberOfAllTexts - numberOfTexts > 0u)
+ return true; // Empty texts exists
+ return false; // No empty texts exists
+ }
TextPosition sp = 0, ep = 0;
// Match including end-marker
/******************************************************************
* Counting queries
*/
+ulong CSA::Count(uchar const * pattern) const
+{
+ TextPosition m = strlen((char *)pattern);
+ if (m == 0)
+ return 0;
+
+ TextPosition sp = 0, ep = 0;
+ unsigned count = (unsigned) Search(pattern, m, &sp, &ep);
+ return count;
+}
+
unsigned CSA::CountPrefix(uchar const * pattern) const
{
TextPosition m = strlen((char *)pattern);
if (m == 0)
- return numberOfTexts;
+ return numberOfAllTexts;
TextPosition sp = 0, ep = 0;
Search(pattern, m, &sp, &ep);
{
TextPosition m = strlen((char *)pattern);
if (m == 0)
- return numberOfTexts;
+ return numberOfAllTexts;
TextPosition sp = 0, ep = 0;
// Search with end-marker
unsigned CSA::CountEqual(uchar const *pattern) const
{
TextPosition m = strlen((char const *)pattern);
-// if (m == 0)
-// return ; // FIXME Test for empty texts.
+ if (m == 0)
+ return numberOfAllTexts - numberOfTexts; // Empty texts.
TextPosition sp = 0, ep = 0;
// Match including end-marker
unsigned CSA::CountContains(uchar const * pattern) const
{
+ TextPosition m = strlen((char const *)pattern);
+ if (m == 0)
+ return numberOfAllTexts; // Total number of texts.
+
// Here counting is as slow as fetching the result set
// because we would have to filter out occ's that fall within same document.
TextCollection::document_result result = Contains(pattern);
{
TextPosition m = strlen((char *)pattern);
if (m == 0)
- return TextCollection::document_result();
+ return TextCollection::document_result(); // FIXME Should return all 1...k
TextPosition sp = 0, ep = 0;
Search(pattern, m, &sp, &ep);
if (resultSize == 0)
return result;
+ result.reserve(resultSize); // Try to avoid reallocation.
+
+ // Iterate through end-markers in [sp,ep]:
unsigned i = 0;
if (sp > 0)
i = alphabetrank->rank(0, sp - 1);
{
// End-marker we found belongs to the "preceeding" doc in the collection
DocId docId = ((*endmarkerDocId)[i] + 1) % numberOfTexts;
+ // Map to doc ID:
+ docId = emptyTextRank->select0(docId+1);
result.push_back(docId);
+
-- resultSize;
++ i;
}
{
TextPosition m = strlen((char *)pattern);
if (m == 0)
- return TextCollection::document_result();
+ return TextCollection::document_result(); // FIXME Should return all 1...k
TextPosition sp = 0, ep = 0;
// Search with end-marker
Search(pattern, m+1, &sp, &ep);
TextCollection::document_result result;
+ result.reserve(ep-sp+1); // Try to avoid reallocation.
+ ulong sampled_rank_i = 0;
// Check each occurrence
for (; sp <= ep; ++sp)
{
TextPosition i = sp;
- uchar c = alphabetrank->charAtPos(i);
- while (c != '\0' && !sampled->IsBitSet(i))
+
+ uchar c = alphabetrank->access(i);
+ while (c != '\0' && !sampled->IsBitSet(i, &sampled_rank_i))
{
- i = C[c]+alphabetrank->rank(c,i)-1; // LF-mapping
- c = alphabetrank->charAtPos(i);
+ i = C[c]+alphabetrank->rank(c,i)-1;
+ c = alphabetrank->access(i);
}
+ // Assert: c == '\0' OR sampled->IsBitSet(i)
+
if (c == '\0')
{
// Rank among the end-markers in BWT
unsigned endmarkerRank = alphabetrank->rank(0, i) - 1;
// End-marker that we found belongs to the "preceeding" doc in collection:
- DocId docId = ((*endmarkerDocId)[endmarkerRank] + 1) % numberOfTexts;
+ DocId docId = ((*endmarkerDocId)[endmarkerRank] + 1) % numberOfTexts;
+ // Map to doc ID:
+ docId = emptyTextRank->select0(docId+1);
result.push_back(docId);
}
else // Sampled position
{
- TextPosition textPos = (*suffixes)[sampled->rank(i)-1];
- result.push_back(DocIdAtTextPos(textPos));
+ DocId docId = (*suffixDocId)[sampled_rank_i-1]; //sampled->rank(i)-1];
+ // Map to doc ID:
+ docId = emptyTextRank->select0(docId+1);
+ result.push_back(docId);
}
}
{
TextPosition m = strlen((char const *)pattern);
if (m == 0)
- return TextCollection::document_result();
+ return TextCollection::document_result(); // FIXME Should return all empty texts
TextPosition sp = 0, ep = 0;
// Match including end-marker
if (resultSize == 0)
return result;
+ result.reserve(resultSize); // Try to avoid reallocation.
+
unsigned i = 0;
if (sp > 0)
i = alphabetrank->rank(0, sp - 1);
while (resultSize)
{
- // End-marker we found belongs to the "previous" doc in the collection
+ // End-marker we found belongs to the "previous" doc in the collection
DocId docId = ((*endmarkerDocId)[i] + 1) % numberOfTexts;
+ // Map to doc ID:
+ docId = emptyTextRank->select0(docId+1);
result.push_back(docId);
+
-- resultSize;
++ i;
}
TextPosition sp = 0, ep = 0;
// Search all occurrences
Search(pattern, m, &sp, &ep);
-
+
// We want unique document indentifiers, using std::set to collect them
std::set<DocId> resultSet;
+ ulong sampled_rank_i = 0;
// Check each occurrence
for (; sp <= ep; ++sp)
{
TextPosition i = sp;
- uchar c = alphabetrank->charAtPos(i);
- while (c != '\0' && !sampled->IsBitSet(i))
+ uchar c = alphabetrank->access(i);
+ while (c != '\0' && !sampled->IsBitSet(i, &sampled_rank_i))
{
i = C[c]+alphabetrank->rank(c,i)-1; // LF-mapping
- c = alphabetrank->charAtPos(i);
+ c = alphabetrank->access(i);
}
if (c == '\0')
{
}
else
{
- TextPosition textPos = (*suffixes)[sampled->rank(i)-1];
- resultSet.insert(DocIdAtTextPos(textPos));
+ DocId di = (*suffixDocId)[sampled_rank_i-1]; //sampled->rank(i)-1];
+ assert((unsigned)di < numberOfTexts);
+ resultSet.insert(di);
}
}
- // Convert std::set to std::vector (TODO better way to construct result vector?)
- TextCollection::document_result result;
- for (std::set<DocId>::iterator it = resultSet.begin(); it != resultSet.end(); ++it)
- result.push_back(*it);
+ // Convert std::set to std::vector
+ TextCollection::document_result result(resultSet.begin(), resultSet.end());
+ // Map to doc ID's
+ for (document_result::iterator it = result.begin(); it != result.end(); ++it)
+ *it = emptyTextRank->select0(*it+1);
return result;
}
{
TextPosition m = strlen((char *)pattern);
if (m == 0)
- return full_result();
+ return full_result(); // FIXME Throw exception?
TextPosition sp = 0, ep = 0;
// Search all occurrences
Search(pattern, m, &sp, &ep);
full_result result;
+ result.reserve(ep-sp+1); // Try to avoid reallocation.
+ ulong sampled_rank_i = 0;
// Report each occurrence
for (; sp <= ep; ++sp)
{
TextPosition i = sp;
TextPosition dist = 0;
- uchar c = alphabetrank->charAtPos(i);
- while (c != '\0' && !sampled->IsBitSet(i))
+ uchar c = alphabetrank->access(i);
+ while (c != '\0' && !sampled->IsBitSet(i, &sampled_rank_i))
{
- i = C[c]+alphabetrank->rank(c,i)-1; // LF-mapping
- c = alphabetrank->charAtPos(i);
+ i = C[c]+alphabetrank->rank(c,i)-1;
+ c = alphabetrank->access(i);
++ dist;
}
if (c == '\0')
// End-marker that we found belongs to the "preceeding" doc in collection:
DocId docId = ((*endmarkerDocId)[endmarkerRank] + 1) % numberOfTexts;
+ // Map to doc ID:
+ docId = emptyTextRank->select0(docId+1);
result.push_back(make_pair(docId, dist));
}
else
{
- TextPosition textPos = (*suffixes)[sampled->rank(i)-1] + dist;
- // Read document identifier and its starting position in text order
- DocId docId = DocIdAtTextPos(textPos);
- result.push_back(make_pair(docId, textPos - (*textStartPos)[docId]));
+ TextPosition textPos = (*suffixes)[sampled_rank_i-1]+dist; //sampled->rank(i)-1] + dist;
+ DocId docId = (*suffixDocId)[sampled_rank_i-1]; //sampled->rank(i)-1];
+// textPos = textPos - (*textStartPos)[docId]; // Offset inside the text
+
+ // Map to doc ID:
+ docId = emptyTextRank->select0(docId+1);
+ result.push_back(make_pair(docId, textPos));
}
}
}
-/**
- * Finds document identifier for given text position
- *
- * Starting text position of the document is stored into second parameter.
- * Binary searching on text starting positions.
- */
-TextCollection::DocId CSA::DocIdAtTextPos(TextPosition i) const
-{
- assert(i < n);
-
- DocId a = 0;
- DocId b = numberOfTexts - 1;
- while (a < b)
- {
- DocId c = a + (b - a)/2;
- if ((*textStartPos)[c] > i)
- b = c - 1;
- else if ((*textStartPos)[c+1] > i)
- return c;
- else
- a = c + 1;
- }
-
- assert(a < (DocId)numberOfTexts);
- assert(i > (*textStartPos)[a]);
- assert(i < (a == (DocId)numberOfTexts - 1 ? n : (*textStartPos)[a+1]));
- return a;
-}
-
-/**
- * Count end-markers in given interval
- */
-unsigned CSA::CountEndmarkers(TextPosition sp, TextPosition ep) const
-{
- if (sp > ep)
- return 0;
-
- ulong ranksp = 0;
- if (sp != 0)
- ranksp = alphabetrank->rank(0, sp - 1);
-
- return alphabetrank->rank(0, ep) - ranksp;
-}
-
/**
* Save index to a file handle
*
* Throws a std::runtime_error exception on i/o error.
* First byte that is saved represents the version number of the save file.
- * In version 1 files, the data fields are:
- * <uchar version> version info;
- * <unsigned s> samplerate;
- * <ulong n> length of the BWT;
- * <ulong bwt> end marker position in BWT;
- * <uchar *> BWT string of length n;
- * <unsigned r> number of texts;
- * <ulong max> length of longest text;
- * <vector textLength>
- * array of <ulong, ulong> pairs.
- *
- * TODO: Save the data structures instead of BWT sequence?
- * TODO: Don't save textLength and textStartPos arrays.
+ * In version 2 files, the data fields are:
+ * uchar versionFlag;
+ TextPosition n;
+ unsigned samplerate;
+ unsigned C[256];
+ TextPosition bwtEndPos;
+ static_sequence * alphabetrank;
+ BSGAP *sampled;
+ BlockArray *suffixes;
+ BlockArray *suffixDocId;
+ unsigned numberOfTexts;
+ unsigned numberOfAllTexts;
+ ulong maxTextLength;
+ BlockArray *endmarkerDocId;
+ BSGAP *emptyTextRank;
*/
void CSA::Save(FILE *file) const
{
- // Saving version 1 data:
- uchar versionFlag = 1;
+ // Saving version info:
if (std::fwrite(&versionFlag, 1, 1, file) != 1)
throw std::runtime_error("CSA::Save(): file write error (version flag).");
+ if (std::fwrite(&(this->n), sizeof(TextPosition), 1, file) != 1)
+ throw std::runtime_error("CSA::Save(): file write error (n).");
if (std::fwrite(&(this->samplerate), sizeof(unsigned), 1, file) != 1)
throw std::runtime_error("CSA::Save(): file write error (samplerate).");
- if (std::fwrite(&(this->n), sizeof(TextPosition), 1, file) != 1)
- throw std::runtime_error("CSA::Save(): file write error (n).");
+ for(ulong i = 0; i < 256; ++i)
+ if (std::fwrite(this->C + i, sizeof(unsigned), 1, file) != 1)
+ throw std::runtime_error("CSA::Save(): file write error (C table).");
if (std::fwrite(&(this->bwtEndPos), sizeof(TextPosition), 1, file) != 1)
throw std::runtime_error("CSA::Save(): file write error (bwt end position).");
- for (ulong offset = 0; offset < n; offset ++)
- {
- uchar c = alphabetrank->charAtPos(offset);
- if (std::fwrite(&c, sizeof(uchar), 1, file) != 1)
- throw std::runtime_error("CSA::Save(): file write error (bwt sequence).");
- }
-
- unsigned r = numberOfTexts;
- if (std::fwrite(&r, sizeof(unsigned), 1, file) != 1)
- throw std::runtime_error("CSA::Save(): file write error (r).");
-
+ alphabetrank->save(file);
+ sampled->Save(file);
+ suffixes->Save(file);
+ suffixDocId->Save(file);
+
+ if (std::fwrite(&(this->numberOfTexts), sizeof(unsigned), 1, file) != 1)
+ throw std::runtime_error("CSA::Save(): file write error (numberOfTexts).");
+ if (std::fwrite(&(this->numberOfAllTexts), sizeof(unsigned), 1, file) != 1)
+ throw std::runtime_error("CSA::Save(): file write error (numberOfAllTexts).");
if (std::fwrite(&(this->maxTextLength), sizeof(ulong), 1, file) != 1)
throw std::runtime_error("CSA::Save(): file write error (maxTextLength).");
-
- for (r = 0; r < numberOfTexts; ++ r)
- {
- if (std::fwrite(&((*textLength)[r]), sizeof(TextPosition), 1, file) != 1)
- throw std::runtime_error("CSA::Save(): file write error (text length).");
- if (std::fwrite(&((*textStartPos)[r]), sizeof(TextPosition), 1, file) != 1)
- throw std::runtime_error("CSA::Save(): file write error (text start).");
- }
+
+ endmarkerDocId->Save(file);
+ emptyTextRank->Save(file);
+ fflush(file);
}
void CSA::Load(FILE *file, unsigned samplerate)
{
// Delete everything
+#ifdef CSA_TEST_BWT
delete dynFMI; dynFMI = 0;
+#endif
delete alphabetrank; alphabetrank = 0;
delete sampled; sampled = 0;
- delete [] suffixes; suffixes = 0;
- delete [] positions; positions = 0;
+ delete suffixes; suffixes = 0;
+ delete suffixDocId; suffixDocId = 0;
delete [] codetable; codetable = 0;
delete endmarkerDocId; endmarkerDocId = 0;
- delete textLength; textLength = 0;
- delete textStartPos; textStartPos = 0;
+ delete emptyTextRank; emptyTextRank = 0;
this->maxTextLength = 0;
this->numberOfTexts = 0;
+ this->numberOfAllTexts = 0;
this->samplerate = samplerate;
this->n = 0;
- uchar versionFlag = 0;
- if (std::fread(&versionFlag, 1, 1, file) != 1)
+ uchar verFlag = 0;
+ if (std::fread(&verFlag, 1, 1, file) != 1)
throw std::runtime_error("CSA::Load(): file read error (version flag).");
- if (versionFlag != 1)
- throw std::runtime_error("CSA::Load(): invalid start byte.");
+ if (verFlag != CSA::versionFlag)
+ throw std::runtime_error("CSA::Load(): invalid save file version.");
+ if (std::fread(&(this->n), sizeof(TextPosition), 1, file) != 1)
+ throw std::runtime_error("CSA::Load(): file read error (n).");
if (std::fread(&samplerate, sizeof(unsigned), 1, file) != 1)
throw std::runtime_error("CSA::Load(): file read error (samplerate).");
- if (this->samplerate == 0)
- this->samplerate = samplerate;
+// FIXME samplerate can not be changed during load.
+// if (this->samplerate == 0)
+// this->samplerate = samplerate;
- if (std::fread(&(this->n), sizeof(TextPosition), 1, file) != 1)
- throw std::runtime_error("CSA::Load(): file read error (n).");
+ for(ulong i = 0; i < 256; ++i)
+ if (std::fread(this->C + i, sizeof(unsigned), 1, file) != 1)
+ throw std::runtime_error("CSA::Load(): file read error (C table).");
if (std::fread(&(this->bwtEndPos), sizeof(TextPosition), 1, file) != 1)
throw std::runtime_error("CSA::Load(): file read error (bwt end position).");
- uchar *bwt = new uchar[n];
- for (ulong offset = 0; offset < n; offset ++)
- if (std::fread((bwt + offset), sizeof(uchar), 1, file) != 1)
- throw std::runtime_error("CSA::Load(): file read error (bwt sequence).");
-
- unsigned r = 0;
- if (std::fread(&r, sizeof(unsigned), 1, file) != 1)
- throw std::runtime_error("CSA::Load(): file read error (r).");
+ alphabetrank = static_sequence::load(file);
+ sampled = new BSGAP(file);
+ suffixes = new BlockArray(file);
+ suffixDocId = new BlockArray(file);
+ if (std::fread(&(this->numberOfTexts), sizeof(unsigned), 1, file) != 1)
+ throw std::runtime_error("CSA::Load(): file read error (numberOfTexts).");
+ if (std::fread(&(this->numberOfAllTexts), sizeof(unsigned), 1, file) != 1)
+ throw std::runtime_error("CSA::Load(): file read error (numberOfAllTexts).");
if (std::fread(&(this->maxTextLength), sizeof(ulong), 1, file) != 1)
throw std::runtime_error("CSA::Load(): file read error (maxTextLength).");
-
- // Build up arrays for text length and starting positions
- textLength = new BlockArray(r, Tools::CeilLog2(maxTextLength));
- textStartPos = new BlockArray(r, Tools::CeilLog2(this->n));
- while (r > 0)
- {
- TextPosition length = 0, start = 0;
- if (std::fread(&length, sizeof(TextPosition), 1, file) != 1)
- throw std::runtime_error("CSA::Load(): file read error (text length).");
- if (std::fread(&start, sizeof(TextPosition), 1, file) != 1)
- throw std::runtime_error("CSA::Load(): file read error (text start).");
-
- (*textLength)[numberOfTexts] = length;
- (*textStartPos)[numberOfTexts] = start;
- ++numberOfTexts;
- --r;
- }
+ endmarkerDocId = new BlockArray(file);
+ emptyTextRank = new BSGAP(file);
- // Construct data structures
- makewavelet(bwt);
- delete [] bwt;
- maketables(); // FIXME: this will redo text length tables
+ // FIXME Construct data structures with new samplerate
+ //maketables();
}
* Rest of the functions follow...
*/
-uchar * CSA::Substring(TextPosition i, TextPosition l) const
-{
- uchar *result = new uchar[l + 1];
- if (l == 0)
- {
- result[0] = 0u;
- return result;
- }
-
- TextPosition dist;
- TextPosition k = i + l - 1;
- // Check for end of the string
- if (k > n - 1)
- {
- l -= k - n + 1;
- k = n - 1;
- }
-
- TextPosition skip = samplerate - k % samplerate - 1;
- TextPosition j;
- if (k / samplerate + 1 >= n / samplerate)
- {
- j = bwtEndPos;
- skip = n - k - 1;
- }
- else
- {
- j = (*positions)[k/samplerate+1];
- //cout << samplerate << ' ' << j << '\n';
- }
-
- for (dist = 0; dist < skip + l; dist++)
- {
- int c = alphabetrank->charAtPos(j);
- if (c == '\0')
- {
- // Rank among the end-markers in BWT
- unsigned endmarkerRank = alphabetrank->rank(0, j) - 1;
- j = (*endmarkerDocId)[endmarkerRank]; // LF-mapping for end-marker
- }
- else
- j = C[c]+alphabetrank->rank(c,j)-1; // LF-mapping
- if (dist >= skip)
- result[l + skip - dist - 1] = c;
- }
- result[l] = 0u;
- return result;
-}
-
- /*TextPosition CSA::inverse(TextPosition i)
-{
- TextPosition skip = samplerate - i % samplerate;
- TextPosition j;
- if (i / samplerate + 1 >= n / samplerate)
- {
- j = bwtEndPos;
- skip = n - i;
- }
- else
- {
- j = (*positions)[i/samplerate+1];
- //cout << samplerate << ' ' << j << '\n';
- }
-
- while (skip > 0)
- {
- int c = alphabetrank->charAtPos(j);
- j = C[c]+alphabetrank->rank(c,j)-1; // LF-mapping
- skip --;
- }
- return j;
- }*/
-ulong CSA::Search(uchar const * pattern, TextPosition m, TextPosition *spResult, TextPosition *epResult) const {
+ulong CSA::Search(uchar const * pattern, TextPosition m, TextPosition *spResult, TextPosition *epResult) const
+{
// use the FM-search replacing function Occ(c,1,i) with alphabetrank->rank(c,i)
int c = (int)pattern[m-1];
TextPosition i=m-1;
return ep - sp + 1;
else
return 0;
- }
-
- /*TextPosition CSA::LF(uchar c, TextPosition &sp, TextPosition &ep)
-{
- sp = C[(int)c]+alphabetrank->rank(c,sp-1);
- ep = C[(int)c]+alphabetrank->rank(c,ep)-1;
-
- if (sp<=ep)
- return ep - sp + 1;
- else
- return 0;
- }*/
-
-CSA::TextPosition CSA::Lookup(TextPosition i) const // Time complexity: O(samplerate log \sigma)
-{
- TextPosition dist=0;
- while (!sampled->IsBitSet(i))
- {
- int c = alphabetrank->charAtPos(i);
- if (c == '\0')
- {
- // End-markers are sampled
- unsigned endmarkerRank = alphabetrank->rank(0, i) - 1;
- DocId docId = (*endmarkerDocId)[endmarkerRank];
- return (*textStartPos)[docId] + dist;
- }
- i = C[c]+alphabetrank->rank(c,i)-1; // LF-mapping
- ++dist;
- }
-
- return (*suffixes)[sampled->rank(i)-1]+dist;
}
CSA::~CSA() {
+#ifdef CSA_TEST_BWT
delete dynFMI;
+#endif
delete alphabetrank;
delete sampled;
delete suffixes;
- delete positions;
- delete [] codetable;
+ delete suffixDocId;
+ delete [] codetable; // FIXME remove
delete endmarkerDocId;
- delete textLength;
- delete textStartPos;
+ delete emptyTextRank;
}
void CSA::makewavelet(uchar *bwt)
for (i=255;i>=min;--i)
if (C[i]>0) {max = i; break;}
- // Print frequencies
- /*for(i = 0; i < 256; i++)
- if (C[i]>0) printf("C[%lu] = %lu\n", i, C[i]);
- fflush(stdout);*/
-
ulong prev=C[0], temp;
C[0]=0;
for (i=1;i<256;i++) {
C[i]=C[i-1]+prev;
prev = temp;
}
- this->codetable = node::makecodetable(bwt,n);
- alphabetrank = new THuffAlphabetRank(bwt,n, this->codetable,0);
- //if (alphabetrank->Test(bwt,n)) printf("alphabetrank ok\n");
-
-/* for (i = 0; i < n; ++i)
- {
- uchar c = alphabetrank->charAtPos(i);
- TextPosition j = C[c]+alphabetrank->rank(c, i)-1;
- printf("LF[%lu] = %lu\n", i, j);
- }*/
+// this->codetable = node::makecodetable(bwt,n);
+// alphabetrank = new THuffAlphabetRank(bwt,n, this->codetable,0);
+// delete [] bwt;
+ //alphabetrank = new RLWaveletTree(bwt, n); // Deletes bwt!
+
+ alphabet_mapper * am = new alphabet_mapper_none();
+ static_bitsequence_builder * bmb = new static_bitsequence_builder_brw32(16); // FIXME samplerate?
+ wt_coder * wtc = new wt_coder_huff(bwt,n,am);
+ alphabetrank = new static_sequence_wvtree(bwt,n,wtc,bmb,am);
+ delete bmb;
+ delete [] bwt;
}
void CSA::maketables()
{
+ // Calculate BWT end-marker position (of last inserted text)
+ // Note: bwtEndPos already set!
+ // and the length of the first text (counter l):
+#ifdef CSA_TEST_BWT
+ ulong i = 0;
+ ulong l = 1;
+
+ uchar c = alphabetrank->access(i);
+ while (c != '\0')
+ {
+ i = C[c]+alphabetrank->rank(c, i)-1;
+ ++l;
+ c = alphabetrank->access(i);
+ }
+ assert(i == bwtEndPos); // compare result
+#endif
+
+ // Build up arrays for text length and starting positions
+ // FIXME Temp, remove
+ //BlockArray* textLength = new BlockArray(numberOfTexts, Tools::CeilLog2(maxTextLength));
+ BlockArray* textStartPos = new BlockArray(numberOfTexts, Tools::CeilLog2(this->n));
+ //(*textLength)[0] = l;
+ (*textStartPos)[0] = 0;
+
+ // Construct samples
ulong sampleLength = (n%samplerate==0) ? n/samplerate : n/samplerate+1;
-
- ulong *sampledpositions = new ulong[n/W+1];
unsigned ceilLog2n = Tools::CeilLog2(n);
- suffixes = new BlockArray(sampleLength, ceilLog2n);
- positions = new BlockArray(sampleLength, ceilLog2n);
-
+ BlockArray* positions = new BlockArray(sampleLength, ceilLog2n);
+ BlockArray* tmpSuffix = new BlockArray(sampleLength, ceilLog2n);
+
// Mapping from end-markers to doc ID's:
endmarkerDocId = new BlockArray(numberOfTexts, Tools::CeilLog2(numberOfTexts));
- ulong i,j=0;
- for (i=0;i<n/W+1;i++)
+ ulong *sampledpositions = new ulong[n/W+1];
+ for (ulong i=0;i<n/W+1;i++)
sampledpositions[i]=0lu;
ulong x,p=bwtEndPos;
- // Keeping track of text position of end-markers seen
+ ulong sampleCount = 0;
+ // Keeping track of text position of prev. end-marker seen
ulong posOfSuccEndmarker = n;
DocId textId = numberOfTexts;
ulong ulongmax = 0;
ulongmax--;
//positions:
- for (i=n-1;i<ulongmax;i--) { // TODO bad solution with ulongmax?
+ for (
ulong i=n-1;i<ulongmax;i--) { // TODO bad solution with ulongmax?
// i substitutes SA->GetPos(i)
x=(i==n-1)?0:i+1;
- if (x % samplerate == 0) {
+ if (x % samplerate == 0 && posOfSuccEndmarker - x > samplerate) {
Tools::SetField(sampledpositions,1,p,1);
- (*positions)[x/samplerate] = p;
+ (*positions)[sampleCount] = p;
+ (*tmpSuffix)[sampleCount] = x; // FIXME remove
+ sampleCount ++;
}
- uchar c = alphabetrank->charAtPos(p);
+ uchar c = alphabetrank->access(p);
if (c == '\0')
{
--textId;
// Store text length and text start position:
if (textId < (DocId)numberOfTexts - 1)
{
- (*textLength)[textId + 1] = posOfSuccEndmarker - x;
+ //(*textLength)[textId + 1] = posOfSuccEndmarker - x;
(*textStartPos)[textId + 1] = x; // x is the position of end-marker.
posOfSuccEndmarker = x;
}
}
assert(textId == 0);
- /*i = 0;
- for (map<ulong, pair<DocId, ulong> >::iterator it = endmarkers.begin(); it != endmarkers.end(); ++it, ++i)
- {
- int docc = (*endmarkerDocId)[i];
- ulong poss = (*endmarkerPos)[i];
- printf("endm[%u] = %lu (text pos: %lu) (recorded: %d, %lu)\n", (it->second).first, it->first, (it->second).second, docc, poss);
- }*/
-/*
- for (i = 0; i < numberOfTexts; ++ i)
+ sampled = new BSGAP(sampledpositions,n,true);
+ sampleLength = sampled->rank(n-1);
+ assert(sampleCount == sampleLength);
+
+ // Suffixes store an offset from the text start position
+ suffixes = new BlockArray(sampleLength, Tools::CeilLog2(maxTextLength));
+ suffixDocId = new BlockArray(sampleLength, Tools::CeilLog2(numberOfTexts));
+
+ for(ulong i=0; i<sampleLength; i++) {
+ assert((*positions)[i] < n);
+ ulong j = sampled->rank((*positions)[i]);
+ if (j==0) j=sampleLength;
+ TextPosition textPos = (*tmpSuffix)[i];
+ (*suffixDocId)[j-1] = DocIdAtTextPos(textStartPos, textPos);
+
+ assert((unsigned)DocIdAtTextPos(textStartPos, textPos) < numberOfTexts);
+ assert((*suffixDocId)[j-1] < numberOfTexts);
+ // calculate offset from text start:
+ (*suffixes)[j-1] = textPos - (*textStartPos)[(*suffixDocId)[j-1]];
+ }
+ // FIXME Temp, remove
+ delete tmpSuffix;
+ delete positions;
+// delete textLength;
+ delete textStartPos;
+}
+
+
+/**
+ * Finds document identifier for given text position
+ *
+ * Starting text position of the document is stored into second parameter.
+ * Binary searching on text starting positions.
+ */
+TextCollection::DocId CSA::DocIdAtTextPos(BlockArray* textStartPos, TextPosition i) const
+{
+ assert(i < n);
+
+ DocId a = 0;
+ DocId b = numberOfTexts - 1;
+ while (a < b)
{
- //std::cout << "textlength = " << dynTextLength[i].first << " vs " << (*textLength)[i] << ", textStartPos = " << dynTextLength[i].second << " vs " << (*textStartPos)[i] << std::endl;
- assert(dynTextLength[i].first == (*textLength)[i]);
- assert(dynTextLength[i].second == (*textStartPos)[i]);
+ DocId c = a + (b - a)/2;
+ if ((*textStartPos)[c] > i)
+ b = c - 1;
+ else if ((*textStartPos)[c+1] > i)
+ return c;
+ else
+ a = c + 1;
}
-*/
- sampled = new BitRank(sampledpositions,n,true);
-
- //suffixes:
- for(i=0; i<sampleLength; i++) {
- j = sampled->rank((*positions)[i]);
- if (j==0) j=sampleLength;
- (*suffixes)[j-1] = (i*samplerate==n)?0:i*samplerate;
- }
+
+ assert(a < (DocId)numberOfTexts);
+ assert(i >= (*textStartPos)[a]);
+ assert(i < (a == (DocId)numberOfTexts - 1 ? n : (*textStartPos)[a+1]));
+ return a;
}
CSA::TCodeEntry * CSA::node::makecodetable(uchar *text, TextPosition n)
return x | (bit << pos);
}
+} // namespace SXSI
+
projects / SXSI / TextCollection.git / blobdiff