// returns a new instance of FieldInfoList which represents field entities in a document index for a specific field, you must delete the instance later. @see FieldInfoList
// Note that not all index types support fields - those that do should override this method.
lemur::api::FieldInfoList *lemur::index::LemurIndriIndex::fieldInfoList(lemur::api::DOCID_T docID, int fieldID) const {
// get the index for this document
indri::collection::Repository::index_state indexes = _repository->indexes();
indri::index::Index* index = _indexWithDocument( indexes, docID );
// ensure we do have an index (i.e. if the docID was invalid...)
if (!index) return NULL;
// and the indri term-list
const indri::index::TermList *tList=index->termList((int)docID);
// ensure we have a term list!
if (!tList) return NULL;
// create a blank field info list object
lemur::api::IndriFieldInfoList *retVal=new lemur::api::IndriFieldInfoList();
// loop through our fields and insert those that match the field ID
const indri::utility::greedy_vector< indri::index::FieldExtent > fieldVec=tList->fields();
int numFields=fieldVec.size();
for (int i=0; i < numFields; i++) {
indri::index::FieldExtent thisField=fieldVec[i];
if (thisField.id==fieldID) {
retVal->add(thisField);
}
}
delete tList;
// and return
return retVal;
}
void indri::index::MemoryIndex::_removeClosedTags( indri::utility::greedy_vector<indri::parse::TagExtent *>& tags, unsigned int position ) {
for( size_t i=0; i<tags.size(); ) {
if( tags[i]->end <= int(position) ) {
tags.erase( tags.begin() + i );
} else {
i++;
}
}
}
void indri::index::MemoryIndex::_writeFieldExtents( lemur::api::DOCID_T documentID, indri::utility::greedy_vector<indri::parse::TagExtent *>& indexedTags ) {
indri::utility::HashTable< indri::parse::TagExtent *, int> tagIdMap;
// sort fields
std::sort( indexedTags.begin(), indexedTags.end(), indri::parse::LessTagExtent() );
// we'll add to the end of the fields greedy_vector
indri::utility::greedy_vector<indri::index::FieldExtent> & fields = _termList.fields();
// this is used to set the parentOrdinals
int offset = fields.size();
// convert to field extents, set ids, and create the node map
for( size_t i=0; i<indexedTags.size(); i++ ) {
indri::parse::TagExtent * extent = indexedTags[i];
int ordinal = int(i) + 1;
// this is the id for the field type
int tagId = _fieldID( extent->name );
// convert the field
indri::index::FieldExtent converted( tagId, extent->begin, extent->end, extent->number, ordinal);
// add this node to the map;
tagIdMap.insert( extent, ordinal );
// add his field to the field list for the document
fields.push_back( converted );
// add this location to the inverted list for fields - deferred to below -
// _fieldLists[tagId - 1]->addLocation( documentID, extent->begin, extent->end, extent->number, ordinal );
}
// set the parent ordinals
for( size_t j=0; j<indexedTags.size(); j++ ) {
indri::parse::TagExtent * extent = indexedTags[j];
// look up the parent
int parentOrdinal = 0;
int * parentIter;
if ( extent->parent != 0 ) {
parentIter = tagIdMap.find( extent->parent );
if( parentIter != 0 ) {
parentOrdinal = *parentIter;
} else {
parentOrdinal = 0;
}
}
// set the parent
int ordinal = fields[ offset + j ].ordinal;
int tagId = fields[ offset + j ].id;
fields[ offset + j ].parentOrdinal = parentOrdinal;
// add this location to the inverted list for fields
_fieldLists[tagId - 1]->addLocation( documentID, extent->begin, extent->end, extent->number, ordinal, parentOrdinal );
}
}
void IndexWriter::_fetchMatchingInvertedLists( indri::utility::greedy_vector<WriterIndexContext*>& lists, invertedlist_pqueue& queue ) {
lists.clear();
WriterIndexContext* first = queue.top();
lists.push_back( first );
const char* firstTerm = first->iterator->currentEntry()->termData->term;
queue.pop();
while( queue.size() && !strcmp( firstTerm, queue.top()->iterator->currentEntry()->termData->term ) ) {
lists.push_back( queue.top() );
queue.pop();
}
}
// Moved work of identifying matches of an extent from ExtentRestriction to here.
// This allows the computation for the list of matches for an extent to be
// overridden. ExtentParent, ExtentChild, and ExtentDescendant are among the
// classes that need to do this, as the parent/child/descedant relationships
// are not based on extent containment - these relationships may be among
// arbitrary fields in a document.
virtual const indri::utility::greedy_vector<indri::index::Extent>& matches( indri::index::Extent &extent ) {
int begin = extent.begin;
int end = extent.end;
const indri::utility::greedy_vector<indri::index::Extent>& exts = extents();
_matches.clear();
for( size_t i = 0 ; i < exts.size(); i++ ) {
if ( begin <= exts[i].begin && end >= exts[i].end ) {
_matches.push_back( exts[i] );
} else if ( exts[i].begin > end ) {
break;
}
}
return _matches;
}
void indri::infnet::FixedPassageNode::_buildSubextents( const indri::utility::greedy_vector<indri::index::Extent>& extents ) {
_subextents.clear();
for( size_t i=0; i<extents.size(); i++ ) {
_addSubextents( extents[i] );
}
}
void IndexWriter::_pushInvertedLists( indri::utility::greedy_vector<WriterIndexContext*>& lists, invertedlist_pqueue& queue ) {
for( int i=0; i<lists.size(); i++ ) {
lists[i]->iterator->nextEntry();
if( !lists[i]->iterator->finished() )
queue.push( lists[i] );
}
}
// Moved work of identifying matches of an extent from ExtentRestriction to here.
// This allows the computation for the list of matches for an extent to be
// overridden. ExtentParent, ExtentChild, and ExtentDescendant are among the
// classes that need to do this, as the parent/child/descedant relationships
// are not based on extent containment - these relationships may be among
// arbitrary fields in a document.
virtual const indri::utility::greedy_vector<indri::index::Extent>& matches( indri::index::Extent &extent ) {
int begin = extent.begin;
int end = extent.end;
_matches.clear();
const indri::utility::greedy_vector<indri::index::Extent>& exts = extents();
// if there's no extents or we have no length - just return
if (begin == end || exts.size()==0) return _matches;
// if we are dealing with child extents, we need to reverse the
// list pointer to the last good position
while((_lastpos > 0) && (exts[_lastpos-1].begin >= begin)){
_lastpos--;
}
// now, we make sure we're in the correct position
// after this loop, _lastpos->begin >= begin
while((_lastpos < exts.size()) && (exts[_lastpos].begin < begin)){
_lastpos++;
}
// for default DocListIteratorNode, any extent: begin+1 == end.
while((_lastpos < exts.size()) && (exts[_lastpos].begin < end)) {
if(exts[_lastpos].end <= end) {
indri::index::Extent ext(exts[_lastpos]);
_matches.push_back(ext);
} // end if(_exts[_lastpos].end<=end)
_lastpos++;
} // end while(_lastpos<_exts.size()&&_exts[_lastpos].begin<end)
/***
*** old method of matching child extents - deprecated
*
* for( size_t i = 0 ; i < exts.size(); i++ ) {
* if ( begin <= exts[i].begin && end >= exts[i].end ) {
* _matches.push_back( exts[i] );
* } else if ( exts[i].begin > end ) {
* break;
* }
* }
**/
return _matches;
}
void indri::index::MemoryIndex::_addOpenTags( indri::utility::greedy_vector<indri::parse::TagExtent *>& indexedTags,
indri::utility::greedy_vector<indri::parse::TagExtent *>& openTags,
indri::utility::greedy_vector<indri::parse::TagExtent *>& extents,
unsigned int& extentIndex,
unsigned int position ) {
for( ; extentIndex < extents.size(); extentIndex++ ) {
indri::parse::TagExtent* extent = extents[extentIndex];
if( extent->begin > (int)position )
break;
int tagId = _fieldID( extent->name );
if( tagId == 0 )
continue;
openTags.push_back( extent );
indexedTags.push_back( extent );
}
}
void IndexWriter::_storeMatchInformation( indri::utility::greedy_vector<WriterIndexContext*>& lists, int sequence, indri::index::TermData* termData, UINT64 startOffset, UINT64 endOffset ) {
bool isFrequent = termData->corpus.totalCount > FREQUENT_TERM_COUNT;
if( isFrequent )
_isFrequentCount++;
for( int i=0; i<lists.size(); i++ ) {
WriterIndexContext* list = lists[i];
indri::index::DiskDocListIterator* iterator = dynamic_cast<DiskDocListIterator*>(lists[i]->iterator->currentEntry()->iterator);
bool isMemoryIndex = (iterator == 0);
bool wasFrequent = (isMemoryIndex || iterator->isFrequent());
if( !wasFrequent )
list->wasInfrequentCount++;
if( wasFrequent )
list->wasFrequentCount++;
list->sequenceCount++;
if( !wasFrequent ) {
if( !isFrequent ) {
// common case--remaining infrequent
assert( sequence - _isFrequentCount - 1 >= 0 );
assert( ((sequence -_isFrequentCount - 1) + _isFrequentCount + 1) <= _corpus.uniqueTerms );
list->bitmap->add( list->wasInfrequentCount - 1, sequence - _isFrequentCount - 1 );
} else if( isFrequent ) {
// becoming frequent
list->newlyFrequent->add( list->wasInfrequentCount - 1, termData->term );
}
}
}
if( isFrequent ) {
indri::index::DiskTermData* diskTermData = disktermdata_create( _fields.size() );
::termdata_merge( diskTermData->termData, termData, _fields.size() );
diskTermData->startOffset = startOffset;
diskTermData->length = endOffset - startOffset;
strcpy( const_cast<char*>(diskTermData->termData->term), termData->term );
_topTerms.push_back( diskTermData );
} else {
indri::index::DiskTermData diskTermData;
diskTermData.termData = termData;
diskTermData.startOffset = startOffset;
diskTermData.length = endOffset - startOffset;
diskTermData.termID = sequence - _topTerms.size();
_storeTermEntry( _infrequentTerms, &diskTermData );
}
}
void IndexWriter::_writeStatistics( indri::utility::greedy_vector<WriterIndexContext*>& lists, indri::index::TermData* termData, UINT64& startOffset ) {
indri::utility::greedy_vector<WriterIndexContext*>::iterator iter;
::termdata_clear( termData, _fields.size() );
// find out what term we're writing
strcpy( const_cast<char*>(termData->term), lists[0]->iterator->currentEntry()->termData->term );
for( iter = lists.begin(); iter != lists.end(); ++iter ) {
indri::index::DocListFileIterator::DocListData* listData = (*iter)->iterator->currentEntry();
::termdata_merge( termData, listData->termData, _fields.size() );
}
_termDataBuffer.clear();
indri::utility::RVLCompressStream stream( _termDataBuffer );
stream << termData->term;
::termdata_compress( stream, termData, _fields.size() );
startOffset = _invertedOutput->tell();
int dataSize = stream.dataSize();
_invertedOutput->write( &dataSize, sizeof(UINT32) );
_invertedOutput->write( stream.data(), stream.dataSize() );
}
/// sorts the extents by the beginning extent position (if available)
void sortbegin(indri::utility::greedy_vector<indri::index::Extent>& extents){
int sorted=0;
int lastbegin = 0; int lastend = extents.size();
while(lastbegin<lastend){
int i=lastbegin;
int end=lastend-1;
lastbegin=lastend;
lastend=i;
for (;i<end;i++){
if (extents[i].begin > extents[i+1].begin){
indri::index::Extent x(extents[i]);
extents[i]=extents[i+1];extents[i+1]=x;
if(lastbegin>i)lastbegin=i;
if(lastend<i+1)lastend=i+1;
}
}
}
}
void indri::infnet::ExtentAndNode::_and( indri::utility::greedy_vector<indri::index::Extent>& out, const indri::utility::greedy_vector<indri::index::Extent>& one, const indri::utility::greedy_vector<indri::index::Extent>& two ) {
indri::utility::greedy_vector<indri::index::Extent>::const_iterator oneIter = one.begin();
indri::utility::greedy_vector<indri::index::Extent>::const_iterator twoIter = two.begin();
out.clear();
indri::index::Extent current;
current.begin = 0;
current.end = 0;
while( oneIter != one.end() && twoIter != two.end() ) {
indri::index::Extent intersection;
// compute the intersection (may be 0 length)
intersection.begin = lemur_compat::max( oneIter->begin, twoIter->begin );
intersection.end = lemur_compat::min( oneIter->end, twoIter->end );
intersection.begin = lemur_compat::min( intersection.begin, intersection.end );
if( current.end < intersection.begin ) {
// if last intersection had non-zero length, put it out in the vector
if( current.begin < current.end )
out.push_back( current );
current = intersection;
} else {
// this intersection touches the last intersection,
// so we'll just put them together
current.end = intersection.end;
}
if( oneIter->end == intersection.end ) {
oneIter++;
}
if( twoIter->end == intersection.end ) {
twoIter++;
}
}
if( current.begin != current.end )
_extents.push_back( current );
}
void IndexWriter::_addInvertedListData( indri::utility::greedy_vector<WriterIndexContext*>& lists, indri::index::TermData* termData, indri::utility::Buffer& listBuffer, UINT64& endOffset ) {
indri::utility::greedy_vector<WriterIndexContext*>::iterator iter;
const int minimumSkip = 1<<12; // 4k
int documentsWritten = 0;
const float topdocsFraction = 0.01f;
bool hasTopdocs = termData->corpus.documentCount > TOPDOCS_DOCUMENT_COUNT;
bool isFrequent = termData->corpus.totalCount > FREQUENT_TERM_COUNT;
int topdocsCount = hasTopdocs ? int(termData->corpus.documentCount * 0.01) : 0;
int topdocsSpace = hasTopdocs ? ((topdocsCount*3*sizeof(UINT32)) + sizeof(int)) : 0;
// write a control byte
char control = (hasTopdocs ? 0x01 : 0) | (isFrequent ? 0x02 : 0);
_invertedOutput->write( &control, 1 );
UINT64 initialPosition = _invertedOutput->tell();
// leave some room for the topdocs list
if( hasTopdocs ) {
_invertedOutput->seek( topdocsSpace + initialPosition );
}
// maintain a list of top documents
std::priority_queue<DocListIterator::TopDocument,
std::vector<DocListIterator::TopDocument>,
DocListIterator::TopDocument::greater> topdocs;
double threshold = 0;
int lastDocument = 0;
int positions = 0;
int docs = 0;
// for each matching list:
for( iter = lists.begin(); iter != lists.end(); ++iter ) {
indri::index::DocListFileIterator::DocListData* listData = (*iter)->iterator->currentEntry();
DocListIterator* iterator = listData->iterator;
Index* index = (*iter)->index;
indri::utility::RVLCompressStream stream( listBuffer );
int listDocs = 0;
int listPositions = 0;
while( !iterator->finished() ) {
// get the latest entry from the list
DocListIterator::DocumentData* documentData = iterator->currentEntry();
// add to document counter
docs++; listDocs++;
// update the topdocs list
if( hasTopdocs ) {
int length = index->documentLength( documentData->document );
int count = documentData->positions.size();
// compute DocListIterator::TopDocument::greater (current, top())
// if false, no reason to insert this entry.
// note that the test is inverted.
// int(length * threshold) <= count is equivalent to
// count/length > topdocs.top().count/topdocs.top().length
// but we use < to force breaking a tie in favor of keeping
// the first seen document.
if( int(length * threshold) < count || topdocs.size() < topdocsCount ) {
// form a topdocs entry for this document
DocListIterator::TopDocument topDocument( documentData->document,
count,
length );
topdocs.push( topDocument );
while( topdocs.size() > topdocsCount )
topdocs.pop();
threshold = topdocs.top().count / double(topdocs.top().length);
}
}
if( listBuffer.position() > minimumSkip ) {
// time to write in a skip
_writeBatch( _invertedOutput, documentData->document, listBuffer.position(), listBuffer );
// delta encode documents by batch
lastDocument = 0;
}
assert( documentData->document > lastDocument );
// write this entry out to the list
stream << documentData->document - lastDocument;
stream << (int) documentData->positions.size();
lastDocument = documentData->document;
int lastPosition = 0;
for( int i=0; i<documentData->positions.size(); i++ ) {
stream << (documentData->positions[i] - lastPosition);
lastPosition = documentData->positions[i];
positions++; listPositions++;
}
iterator->nextEntry();
}
indri::index::TermData* td = iterator->termData();
assert( listPositions == td->corpus.totalCount );
assert( listDocs == td->corpus.documentCount );
}
assert( docs == termData->corpus.documentCount );
assert( positions == termData->corpus.totalCount );
// write in the final skip info
_writeBatch( _invertedOutput, -1, listBuffer.position(), listBuffer );
UINT64 finalPosition = _invertedOutput->tell();
if( hasTopdocs ) {
_invertedOutput->seek( initialPosition );
_invertedOutput->write( &topdocsCount, sizeof(int) );
assert( topdocs.size() == topdocsCount );
// write these into the topdocs list in order from smallest fraction to largest fraction,
// where fraction = c(w;D)/|D|
while( topdocs.size() ) {
DocListIterator::TopDocument topDocument = topdocs.top();
_invertedOutput->write( &topDocument.document, sizeof(int) );
_invertedOutput->write( &topDocument.count, sizeof(int) );
_invertedOutput->write( &topDocument.length, sizeof(int) );
topdocs.pop();
}
assert( (_invertedOutput->tell() - initialPosition) == topdocsSpace );
_invertedOutput->seek( finalPosition );
}
endOffset = finalPosition;
}
void _fetchText( indri::utility::greedy_vector<TagExtent *>& tags, indri::utility::greedy_vector<char*>& terms ) {
// now, fetch the additional terms
char line[65536];
_buffer.clear();
for( int i=0; i<_count; i++ ) {
// LINK
_in.getline( line, sizeof line-1 );
// LINKDOCNO
_in.getline( line, sizeof line-1 );
// TEXT=
_in.getline( line, sizeof line-1 );
int textLen = strlen(line+6);
strcpy( _buffer.write(textLen+1), line+6 );
_buffer.unwrite(1);
assert( *(_buffer.front()+_buffer.position()-1) == '\"' && "Last character should be a quote" );
}
*(_buffer.write(1)) = 0;
// now there's a bunch of text in _buffer, space separated, with each
// link separated by a " symbol
char* beginWord = 0;
int beginIndex = 0;
char* buffer = _buffer.front();
for( unsigned int i=0; i<_buffer.position(); i++ ) {
if( isalnum(buffer[i]) && !beginWord ) {
beginWord = buffer+i;
if(!beginIndex)
beginIndex = terms.size();
} else if( isspace(buffer[i]) ) {
buffer[i] = 0;
if( beginWord )
terms.push_back( beginWord );
beginWord = 0;
} else if( buffer[i] == '\"' ) {
buffer[i] = 0;
if( beginWord )
terms.push_back( beginWord );
beginWord = 0;
TagExtent * extent = new TagExtent;
extent->name = "inlink";
extent->begin = beginIndex;
extent->end = terms.size();
extent->number = 0;
extent->parent = 0;
assert( extent->begin <= extent->end );
if( beginIndex ) {
tags.push_back(extent);
if( terms.size() > 125000 )
break;
}
beginIndex = 0;
}
}
}
