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 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::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::_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() );
}
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;
}