src.it.unimi.dsi.big.mg4j.search.ConsecutiveDocumentIterator Maven / Gradle / Ivy
Show all versions of mg4j-big Show documentation
package it.unimi.dsi.big.mg4j.search;
/*
* MG4J: Managing Gigabytes for Java (big)
*
* Copyright (C) 2003-2011 Paolo Boldi and Sebastiano Vigna
*
* This library is free software; you can redistribute it and/or modify it
* under the terms of the GNU Lesser General Public License as published by the Free
* Software Foundation; either version 3 of the License, or (at your option)
* any later version.
*
* This library is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
* or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License
* for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with this program; if not, see As an additional service, this class makes it possible to specify gaps between
* intervals. If gaps are specified, a match will satisfy the condition
* that the left extreme of the first interval is larger than or equal to the
* first gap, the left extreme of the second interval is equal to
* the right extreme of the first interval plus the second gap plus one,
* the left extreme of the third interval is equal to the right extreme
* of the second interval plus the third gap plus one and so on. The standard
* semantics corresponds thus to the everywhere zero gap array. That
* the returned intervals will contain the leftmost gap, too.
*
*
This semantics
* makes it possible to perform phrasal searches “with holes”, typically
* because of stopwords that have not been indexed. Note that it is possible to specify
* a gap before the first interval, but not after the last interval,
* as in general the document length is not known at this level of query resolution.
*
*
This class will handle correctly {@link IntervalIterators#TRUE TRUE} iterators; in this
* case, the semantics is defined as follows: an interval is in the output if it is formed by the union of disjoint intervals,
* one from each input list, and each gap of value k corresponds to k iterators
* returning all document positions as singleton intervals. Since {@link IntervalIterators#TRUE TRUE} represents a list containing just
* the empty interval, the result is equivalent to dropping {@link IntervalIterators#TRUE TRUE} iterators from the input; as
* a consequence, the gap of a {@link IntervalIterators#TRUE TRUE} iterator is merged with that of the following iterator.
*
*
Warning: In case gaps are specified, the mathematically correct semantics would require that
* gaps before {@link IntervalIterators#TRUE TRUE} iterators that are not followed by any non-{@link IntervalIterators#TRUE TRUE} iterators
* have the effect of enlarging the resulting intervals on the right side. However,
* this behaviour is very difficult to implement at this level because document lengths are not known. For this
* reason, if one or more {@link IntervalIterators#TRUE TRUE} iterators appear a the end of the component iterator list they will be simply dropped.
*/
public class ConsecutiveDocumentIterator extends AbstractOrderedIntervalDocumentIterator {
/** The gap array. This is essentially the array provided at construction time; however, if a {@link ConsecutiveIndexIntervalIterator}
* is requested by {@link #getComposedIntervalIterator(Index)} this array will be used to store cumulative gaps. */
private final int gap[];
/** Returns a document iterator that computes the consecutive AND of the given array of iterators.
*
*
Note that the special case of the empty and of the singleton arrays
* are handled efficiently.
*
* @param index the default index; relevant only if it has zero length.
* @param documentIterator the iterators to be composed.
* @return a document iterator that computes the consecutive AND of it.
* @throws IOException
*/
public static DocumentIterator getInstance( final Index index, final DocumentIterator... documentIterator ) throws IOException {
if ( documentIterator.length == 0 ) throw new IllegalArgumentException( "The provided array of document iterators is empty." );
if ( documentIterator.length == 1 ) return documentIterator[ 0 ];
return new ConsecutiveDocumentIterator( documentIterator, null );
}
/** Returns a document iterator that computes the consecutive AND of the given nonzero-length array of iterators.
*
*
Note that the special case of the singleton array is handled efficiently.
*
* @param documentIterator the iterators to be composed (at least one).
* @return a document iterator that computes the consecutive AND of documentIterator.
* @throws IOException
*/
public static DocumentIterator getInstance( final DocumentIterator... documentIterator ) throws IOException {
if ( documentIterator.length == 0 ) throw new IllegalArgumentException( "The provided array of document iterators is empty." );
if ( documentIterator.length == 1 ) return documentIterator[ 0 ];
return getInstance( null, documentIterator );
}
/** Returns a document iterator that computes the consecutive AND of the given nonzero-length array of iterators, adding
* gaps between intervals.
*
*
A match will satisfy the condition
* that the left extreme of the first interval is larger than or equal to the
* first gap, the left extreme of the second interval is larger than
* the right extreme of the first interval plus the second gap, and so on. This semantics
* makes it possible to perform phrasal searches “with holes”, typically
* because of stopwords that have not been indexed.
*
* @param documentIterator the iterators to be composed (at least one).
* @param gap an array of gaps parallel to documentIterator, or null for no gaps.
* @return a document iterator that computes the consecutive AND of documentIterator using the given gaps.
* @throws IOException
*/
public static DocumentIterator getInstance( final DocumentIterator documentIterator[], final int gap[] ) throws IOException {
if ( gap != null && gap.length != documentIterator.length ) throw new IllegalArgumentException( "The number of gaps (" + gap.length + ") is not equal to the number of document iterators (" + documentIterator.length +")" );
if ( documentIterator.length == 1 && ( gap == null || gap[ 0 ] == 0 ) ) return documentIterator[ 0 ];
return new ConsecutiveDocumentIterator( documentIterator, gap );
}
protected ConsecutiveDocumentIterator( final DocumentIterator[] documentIterator, final int[] gap ) throws IOException {
super( documentIterator );
if ( gap == null ) this.gap = new int[ n ];
else this.gap = gap.clone();
}
protected IntervalIterator getComposedIntervalIterator( final Index unused ) {
if ( ASSERTS ) assert unused == soleIndex;
if ( indexIterator == null ) return new ConsecutiveIntervalIterator( gap );
// In this case, gap must be made cumulative
for( int i = 1; i < n; i++ ) gap[ i ] += gap[ i - 1 ] + 1;
return new ConsecutiveIndexIntervalIterator( gap );
}
/** An interval iterator returning the BLOCK of the component iterator
* (i.e., intervals made of sequences of consecutive intervals
* of the component iterator, in the given order).
*
*
In this implementation, {@link #advanced} is
* never true when {@link AbstractOrderedIntervalIterator#endOfProcess} is true.
*/
private class ConsecutiveIntervalIterator extends AbstractOrderedIntervalIterator {
/** A cached reference to the gap array. */
@SuppressWarnings("hiding")
private final int[] gap;
/** The actual gaps. They depend on whether some {@link IntervalIterators#TRUE} iterator reduces the iterator array. */
private final int[] actualGap;
/** Whether the scan is over. */
private boolean endOfProcess;
/** The number of non-{@link IntervalIterators#TRUE} interval iterator. */
private int m;
public ConsecutiveIntervalIterator( final int[] gap ) {
this.gap = gap;
// The enlargement is made necessary by the filling long in reset().
this.actualGap = new int[ n + 1 ];
}
/** Loads {@link #curr} with the first interval from each non-{@link IntervalIterators#TRUE} iterator, leaving
* in {@link #m} the number of non-{@link IntervalIterators#TRUE} iterators, and in {@link #actualGap}
* the gaps to be used for those {@link #m} iterators.
*/
public void reset() throws IOException {
final int[] actualGap = this.actualGap;
final int[] gap = this.gap;
final IntervalIterator[] intervalIterator = this.intervalIterator;
actualGap[ m = 0 ] = -1; // The first interval has actual gap zero if it has gap zero, so we compensate here for the increment below.
int i;
for( i = 0; i < n; i++ ) {
actualGap[ m ] += gap[ i ]; // Accumulate gap
if ( ( intervalIterator[ m ] = documentIterator[ i ].intervalIterator() ) != IntervalIterators.TRUE ) {
actualGap[ m ]++; // If this interval iterator is real, add one.
curr[ m ] = Intervals.MINUS_INFINITY;
actualGap[ ++m ] = 0; // Prepare next gap.
}
}
if ( m == 0 ) throw new IllegalStateException();
next = null;
do {
curr[ 0 ] = intervalIterator[ 0 ].nextInterval();
} while( curr[ 0 ] != null && curr[ 0 ].left < actualGap[ 0 ] );
if ( ! ( endOfProcess = curr[ 0 ] == null ) ) next = align();
}
public void intervalTerms( final LongSet terms ) {
for( int i = m; i-- != 0; ) intervalIterator[ i ].intervalTerms( terms );
}
private Interval align() throws IOException {
if ( DEBUG ) System.err.println( this + ".align()" );
final Interval[] curr = this.curr;
final int[] actualGap = this.actualGap;
final IntervalIterator[] intervalIterator = this.intervalIterator;
if ( DEBUG ) System.err.println( java.util.Arrays.asList( curr ) );
int k = 0;
while( k < m ) {
for ( k = 1; k < m; k++ ) {
while ( curr[ k ].left < curr[ k - 1 ].right + actualGap[ k ] )
if ( ( curr[ k ] = intervalIterator[ k ].nextInterval() ) == null ) {
endOfProcess = true;
return null;
}
if ( curr[ k ].left > curr[ k - 1 ].right + actualGap[ k ] ) {
if ( endOfProcess = ( ( curr[ 0 ] = intervalIterator[ 0 ].nextInterval() ) == null ) ) return null;
break;
}
}
}
return Interval.valueOf( curr[ 0 ].left - actualGap[ 0 ], curr[ m - 1 ].right );
}
public Interval nextInterval() throws IOException {
if ( next != null ) {
final Interval result = next;
next = null;
return result;
}
if ( endOfProcess ) return null;
if ( ( curr[ 0 ] = intervalIterator[ 0 ].nextInterval() ) == null ) {
endOfProcess = true;
return null;
}
return align();
}
public int extent() {
int s = 0;
for ( int i = m; i-- != 0; ) s += intervalIterator[ i ].extent() + actualGap[ i ];
return s - m + 1;
}
}
private class ConsecutiveIndexIntervalIterator extends AbstractOrderedIndexIntervalIterator {
/** A cached reference to the gap array. */
@SuppressWarnings("hiding")
private final int[] gap;
/** Whether the scan is over. */
private boolean endOfProcess;
public ConsecutiveIndexIntervalIterator( final int[] gap ) {
this.gap = gap;
}
/** Resets the iterator by calling the superclass method, and then aligning all iterators.
*
*
Note that in this class {@link #curr} is used to denote the value of the current position
* minus the corresponding {@linkplain #gap cumulative gap}; this method updates {@link #curr} accordingly. */
public void reset() throws IOException {
final int[][] position = this.position;
final int[] currPos = this.currPos;
final int[] curr = this.curr;
final int[] count = this.count;
final int[] gap = this.gap;
IntArrays.fill( currPos, 0 );
for( int i = n; i-- != 0; ) {
count[ i ] = indexIterator[ i ].count();
position[ i ] = indexIterator[ i ].positionArray();
curr[ i ] = position[ i ][ 0 ];
}
endOfProcess = false;
for( int i = n; i-- != 0; ) curr[ i ] -= gap[ i ];
if ( gap[ 0 ] != 0 ) {
// Go beyond the 0-th gap. This must be done just once.
next = null;
while ( curr[ 0 ] < 0 && ++currPos[ 0 ] < count[ 0 ] ) curr[ 0 ] = position[ 0 ][ currPos[ 0 ] ] - gap[ 0 ];
endOfProcess = currPos[ 0 ] == count[ 0 ];
}
if ( ! endOfProcess ) next = align();
}
public void intervalTerms( final LongSet terms ) {
for( int i = n; i-- != 0; ) terms.add( indexIterator[ i ].termNumber() );
}
private Interval align() {
if ( DEBUG ) System.err.println( this + ".align()" );
if ( n == 1 ) return Interval.valueOf( curr[ 0 ], curr[ 0 ] + gap[ 0 ] );
final int[][] position = this.position;
final int[] currPos = this.currPos;
final int[] curr = this.curr;
final int[] count = this.count;
final int[] gap = this.gap;
int c, k = 1, l = n <= 2 ? 0 : 2; // This is actually 2 % n
boolean oneRoundDone = false;
int[] p;
int start = curr[ 0 ];
for(;;) {
p = position[ k ];
c = currPos[ k ];
// First, we try to align the k-th term.
while ( c < count[ k ] && p[ c ] - gap[ k ] < start ) c++;
// If we exhaust the term positions, it's all over.
if ( c == count[ k ] ) {
endOfProcess = true;
return null;
}
curr[ k ] = p[ currPos[ k ] = c ] - gap[ k ];
// If we went beyond start + k, we must update start.
if ( curr[ k ] > start ) start = curr[ k ];
// If k == 0, it means we have made a full round of alignment, so the next check is now valid.
oneRoundDone |= ( k == 0 );
// If oneRoundDone, all current normalised positions (curr[ x ] - gap[ x ]) are squeezed between start and curr[ l ].
if ( oneRoundDone && curr[ l ] == start ) return Interval.valueOf( curr[ 0 ], curr[ 0 ] + gap[ n - 1 ] );
k = l;
if ( ( l = l + 1 ) == n ) l = 0;
}
}
public Interval nextInterval() {
if ( next != null ) {
final Interval result = next;
next = null;
return result;
}
if ( endOfProcess ) return null;
if ( ++currPos[ 0 ] < count[ 0 ] ) curr[ 0 ] = position[ 0 ][ currPos[ 0 ] ] - gap[ 0 ];
else {
endOfProcess = true;
return null;
}
return align();
}
public int extent() {
return gap[ n - 1 ] + 1;
}
}
}