/ Check-in [71d32f53]
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:Add tests and fixes for bm25() function.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | fts5
Files: files | file ages | folders
SHA1: 71d32f53e81921e43c933cc968cb1c18d83fe1e0
User & Date: dan 2014-07-26 18:38:51
Context
2014-07-28
20:14
Add the "loadfts" program, for performance testing the loading of data into fts3/fts4/fts5 tables. check-in: 770b9540 user: dan tags: fts5
2014-07-26
18:38
Add tests and fixes for bm25() function. check-in: 71d32f53 user: dan tags: fts5
2014-07-25
20:30
Add extension apis xRowCount, xQueryPhrase, xSetAuxdata and xGetAuxdata. And a ranking function that uses all of the above. check-in: c4d50428 user: dan tags: fts5
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to ext/fts5/fts5_aux.c.

407
408
409
410
411
412
413
414





415
416
417
418



419
420



421
422
423
424
425
426
427
428
429
430
431
432
433
...
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
...
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512

513
514
515
516
517
518
519
...
526
527
528
529
530
531
532


533
534
535
536
537

538
539
540
541
542
543
544
545
546
547
548
549
550














551


552
553

554
555
556
557
558
559
560
561
562

563
564
565





























































566






567
568





569
570
571
572
573
574
575

576
577
578

579
580

581
582
583
584



585
586

587
588
589
590
591
592



593
594
595

596
597
598
599
600
601
602
...
848
849
850
851
852
853
854

855
856
857
858
859
860
861

  fts5SnippetIterFree(pIter);
  if( rc!=SQLITE_OK ){
    sqlite3_result_error_code(pCtx, rc);
  }
}

typedef struct Fts5GatherCtx Fts5GatherCtx;





struct Fts5GatherCtx {
  int nCol;
  int iPhrase;
  int *anVal;



};




static int fts5GatherCallback(
  const Fts5ExtensionApi *pApi, 
  Fts5Context *pFts,
  void *pUserData
){
  Fts5GatherCtx *p = (Fts5GatherCtx*)pUserData;
  int i = 0;
  int iPrev = -1;
  i64 iPos = 0;

  while( 0==pApi->xPoslist(pFts, 0, &i, &iPos) ){
    int iCol = FTS5_POS2COLUMN(iPos);
    if( iCol!=iPrev ){
................................................................................
){
  int rc = SQLITE_OK;
  int *anVal = 0;
  int i;                          /* For iterating through expression phrases */
  int nPhrase = pApi->xPhraseCount(pFts);
  int nCol = pApi->xColumnCount(pFts);
  int nByte = nCol * nPhrase * sizeof(int);
  Fts5GatherCtx sCtx;

  sCtx.nCol = nCol;
  anVal = sCtx.anVal = (int*)sqlite3_malloc(nByte);
  if( anVal==0 ){
    rc = SQLITE_NOMEM;
  }else{
    memset(anVal, 0, nByte);
................................................................................

  *panVal = anVal;
  return rc;
}

typedef struct Fts5Bm25Context Fts5Bm25Context;
struct Fts5Bm25Context {
  int nPhrase;
  int nCol;
  double *aIDF;                   /* Array of IDF values */
  double *aAvg;                   /* Average size of each column in tokens */
};

static void fts5Bm25Function(
  const Fts5ExtensionApi *pApi,   /* API offered by current FTS version */
  Fts5Context *pFts,              /* First arg to pass to pApi functions */
  sqlite3_context *pCtx,          /* Context for returning result/error */
  int nVal,                       /* Number of values in apVal[] array */
  sqlite3_value **apVal           /* Array of trailing arguments */
){
  const double k1 = 1.2;
  const double B = 0.75;

  int rc = SQLITE_OK;
  Fts5Bm25Context *p;


  p = pApi->xGetAuxdata(pFts, 0);
  if( p==0 ){
    int *anVal = 0;
    int ic;                       /* For iterating through columns */
    int ip;                       /* For iterating through phrases */
    i64 nRow;                     /* Total number of rows in table */
................................................................................
    p = (Fts5Bm25Context*)sqlite3_malloc(nByte);
    if( p==0 ){
      rc = SQLITE_NOMEM;
    }else{
      memset(p, 0, nByte);
      p->aAvg = (double*)&p[1];
      p->aIDF = (double*)&p->aAvg[nCol];


    }

    if( rc==SQLITE_OK ){
      rc = pApi->xRowCount(pFts, &nRow); 
      assert( nRow>0 || rc!=SQLITE_OK );

    }

    for(ic=0; rc==SQLITE_OK && ic<nCol; ic++){
      i64 nToken = 0;
      rc = pApi->xColumnTotalSize(pFts, ic, &nToken);
      p->aAvg[ic] = (double)nToken / (double)nRow;
    }

    if( rc==SQLITE_OK ){
      rc = fts5GatherTotals(pApi, pFts, &anVal);
    }
    for(ic=0; ic<nCol; ic++){
      for(ip=0; rc==SQLITE_OK && ip<nPhrase; ip++){














        int idx = ip * nCol + ic;


        p->aIDF[idx] = log( (0.5 + nRow - anVal[idx]) / (0.5 + anVal[idx]) );
        if( p->aIDF[idx]<0.0 ) p->aIDF[idx] = 0.0;

      }
    }

    sqlite3_free(anVal);
    if( rc==SQLITE_OK ){
      rc = pApi->xSetAuxdata(pFts, p, sqlite3_free);
    }
    if( rc!=SQLITE_OK ){
      sqlite3_free(p);

    }
  }






























































  if( rc==SQLITE_OK ){






    int ip;
    double score = 0.0;






    for(ip=0; rc==SQLITE_OK && ip<p->nPhrase; ip++){
      int iPrev = 0;
      int nHit = 0;
      int i = 0;
      i64 iPos = 0;


      while( rc==SQLITE_OK && 0==pApi->xPoslist(pFts, ip, &i, &iPos) ){
        int iCol = FTS5_POS2COLUMN(iPos);
        if( iCol!=iPrev && nHit>0 ){

          int sz = 0;
          int idx = ip * p->nCol + iPrev;

          rc = pApi->xColumnSize(pFts, iPrev, &sz);

          score += p->aIDF[idx] * nHit * (k1+1.0) / 
                  (nHit + k1 * (1.0 - B + B * sz / p->aAvg[iCol]));



          nHit = 0;
        }

        nHit++;
        iPrev = iCol;
      }
    }
    
    if( rc==SQLITE_OK ){



      sqlite3_result_double(pCtx, score);
    }


  }

  if( rc!=SQLITE_OK ){
    sqlite3_result_error_code(pCtx, rc);
  }
}

................................................................................
  struct Builtin {
    const char *zFunc;            /* Function name (nul-terminated) */
    void *pUserData;              /* User-data pointer */
    fts5_extension_function xFunc;/* Callback function */
    void (*xDestroy)(void*);      /* Destructor function */
  } aBuiltin [] = {
    { "bm25",      0, fts5Bm25Function,    0 },

    { "snippet",   0, fts5SnippetFunction, 0 },
    { "fts5_test", 0, fts5TestFunction,    0 },
  };

  int rc = SQLITE_OK;             /* Return code */
  int i;                          /* To iterate through builtin functions */








<
>
>
>
>
>

<
<
<
>
>
>


>
>
>



|

|







 







|







 







|
|




|


|
<
<

<
<
<
<

>







 







>
>





>













>
>
>
>
>
>
>
>
>
>
>
>
>
>
|
>
>
|
|
>









>



>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>

>
>
>
>
>
>


>
>
>
>
>







>
|

<
>


>


|
|
>
>
>


>




|

>
>
>
|
|
|
>







 







>







407
408
409
410
411
412
413

414
415
416
417
418
419



420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
...
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
...
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511


512




513
514
515
516
517
518
519
520
521
...
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673

674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
...
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967

  fts5SnippetIterFree(pIter);
  if( rc!=SQLITE_OK ){
    sqlite3_result_error_code(pCtx, rc);
  }
}



/*
** Context object passed by fts5GatherTotals() to xQueryPhrase callback
** fts5GatherCallback().
*/
struct Fts5GatherCtx {



  int nCol;                       /* Number of columns in FTS table */
  int iPhrase;                    /* Phrase currently under investigation */
  int *anVal;                     /* Array to populate */
};

/*
** Callback used by fts5GatherTotals() with the xQueryPhrase() API.
*/
static int fts5GatherCallback(
  const Fts5ExtensionApi *pApi, 
  Fts5Context *pFts,
  void *pUserData                 /* Pointer to Fts5GatherCtx object */
){
  struct Fts5GatherCtx *p = (struct Fts5GatherCtx*)pUserData;
  int i = 0;
  int iPrev = -1;
  i64 iPos = 0;

  while( 0==pApi->xPoslist(pFts, 0, &i, &iPos) ){
    int iCol = FTS5_POS2COLUMN(iPos);
    if( iCol!=iPrev ){
................................................................................
){
  int rc = SQLITE_OK;
  int *anVal = 0;
  int i;                          /* For iterating through expression phrases */
  int nPhrase = pApi->xPhraseCount(pFts);
  int nCol = pApi->xColumnCount(pFts);
  int nByte = nCol * nPhrase * sizeof(int);
  struct Fts5GatherCtx sCtx;

  sCtx.nCol = nCol;
  anVal = sCtx.anVal = (int*)sqlite3_malloc(nByte);
  if( anVal==0 ){
    rc = SQLITE_NOMEM;
  }else{
    memset(anVal, 0, nByte);
................................................................................

  *panVal = anVal;
  return rc;
}

typedef struct Fts5Bm25Context Fts5Bm25Context;
struct Fts5Bm25Context {
  int nPhrase;                    /* Number of phrases in query */
  int nCol;                       /* Number of columns in FTS table */
  double *aIDF;                   /* Array of IDF values */
  double *aAvg;                   /* Average size of each column in tokens */
};

static int fts5Bm25GetContext(
  const Fts5ExtensionApi *pApi,   /* API offered by current FTS version */
  Fts5Context *pFts,              /* First arg to pass to pApi functions */
  Fts5Bm25Context **pp            /* OUT: Context object */


){




  Fts5Bm25Context *p;
  int rc = SQLITE_OK;

  p = pApi->xGetAuxdata(pFts, 0);
  if( p==0 ){
    int *anVal = 0;
    int ic;                       /* For iterating through columns */
    int ip;                       /* For iterating through phrases */
    i64 nRow;                     /* Total number of rows in table */
................................................................................
    p = (Fts5Bm25Context*)sqlite3_malloc(nByte);
    if( p==0 ){
      rc = SQLITE_NOMEM;
    }else{
      memset(p, 0, nByte);
      p->aAvg = (double*)&p[1];
      p->aIDF = (double*)&p->aAvg[nCol];
      p->nCol = nCol;
      p->nPhrase = nPhrase;
    }

    if( rc==SQLITE_OK ){
      rc = pApi->xRowCount(pFts, &nRow); 
      assert( nRow>0 || rc!=SQLITE_OK );
      if( nRow<2 ) nRow = 2;
    }

    for(ic=0; rc==SQLITE_OK && ic<nCol; ic++){
      i64 nToken = 0;
      rc = pApi->xColumnTotalSize(pFts, ic, &nToken);
      p->aAvg[ic] = (double)nToken / (double)nRow;
    }

    if( rc==SQLITE_OK ){
      rc = fts5GatherTotals(pApi, pFts, &anVal);
    }
    for(ic=0; ic<nCol; ic++){
      for(ip=0; rc==SQLITE_OK && ip<nPhrase; ip++){
        /* Calculate the IDF (Inverse Document Frequency) for phrase ip
        ** in column ic. This is done using the standard BM25 formula as
        ** found on wikipedia:
        **
        **   IDF = log( (N - nHit + 0.5) / (nHit + 0.5) )
        **
        ** where "N" is the total number of documents in the set and nHit
        ** is the number that contain at least one instance of the phrase
        ** under consideration.
        **
        ** The problem with this is that if (N < 2*nHit), the IDF is 
        ** negative. Which is undesirable. So the mimimum allowable IDF is
        ** (1e-6) - roughly the same as a term that appears in just over
        ** half of set of 5,000,000 documents.  */
        int idx = ip * nCol + ic; /* Index in aIDF[] and anVal[] arrays */
        int nHit = anVal[idx];    /* Number of docs matching "ic: ip" */

        p->aIDF[idx] = log( (0.5 + nRow - nHit) / (0.5 + nHit) );
        if( p->aIDF[idx]<=0.0 ) p->aIDF[idx] = 1e-6;
        assert( p->aIDF[idx]>=0.0 );
      }
    }

    sqlite3_free(anVal);
    if( rc==SQLITE_OK ){
      rc = pApi->xSetAuxdata(pFts, p, sqlite3_free);
    }
    if( rc!=SQLITE_OK ){
      sqlite3_free(p);
      p = 0;
    }
  }

  *pp = p;
  return rc;
}

static void fts5Bm25DebugContext(
  int *pRc,                       /* IN/OUT: Return code */
  Fts5Buffer *pBuf,               /* Buffer to populate */
  Fts5Bm25Context *p              /* Context object to decode */
){
  int ip;
  int ic;

  sqlite3Fts5BufferAppendString(pRc, pBuf, "idf ");
  if( p->nPhrase>1 || p->nCol>1 ){
    sqlite3Fts5BufferAppendString(pRc, pBuf, "{");
  }
  for(ip=0; ip<p->nPhrase; ip++){
    if( ip>0 ) sqlite3Fts5BufferAppendString(pRc, pBuf, " ");
    if( p->nCol>1 ) sqlite3Fts5BufferAppendString(pRc, pBuf, "{");
    for(ic=0; ic<p->nCol; ic++){
      if( ic>0 ) sqlite3Fts5BufferAppendString(pRc, pBuf, " ");
      sqlite3Fts5BufferAppendPrintf(pRc, pBuf, "%f", p->aIDF[ip*p->nCol+ic]);
    }
    if( p->nCol>1 ) sqlite3Fts5BufferAppendString(pRc, pBuf, "}");
  }
  if( p->nPhrase>1 || p->nCol>1 ){
    sqlite3Fts5BufferAppendString(pRc, pBuf, "}");
  }

  sqlite3Fts5BufferAppendString(pRc, pBuf, " avgdl ");
  if( p->nCol>1 ) sqlite3Fts5BufferAppendString(pRc, pBuf, "{");
  for(ic=0; ic<p->nCol; ic++){
    if( ic>0 ) sqlite3Fts5BufferAppendString(pRc, pBuf, " ");
    sqlite3Fts5BufferAppendPrintf(pRc, pBuf, "%f", p->aAvg[ic]);
  }
  if( p->nCol>1 ) sqlite3Fts5BufferAppendString(pRc, pBuf, "}");
}

static void fts5Bm25DebugRow(
  int *pRc, 
  Fts5Buffer *pBuf, 
  Fts5Bm25Context *p, 
  const Fts5ExtensionApi *pApi, 
  Fts5Context *pFts
){
}

static void fts5Bm25Function(
  const Fts5ExtensionApi *pApi,   /* API offered by current FTS version */
  Fts5Context *pFts,              /* First arg to pass to pApi functions */
  sqlite3_context *pCtx,          /* Context for returning result/error */
  int nVal,                       /* Number of values in apVal[] array */
  sqlite3_value **apVal           /* Array of trailing arguments */
){
  const double k1 = 1.2;
  const double B = 0.75;
  int rc = SQLITE_OK;
  Fts5Bm25Context *p;

  rc = fts5Bm25GetContext(pApi, pFts, &p);

  if( rc==SQLITE_OK ){
    /* If the bDebug flag is set, instead of returning a numeric rank, this
    ** function returns a text value showing how the rank is calculated. */
    Fts5Buffer debug;
    int bDebug = (pApi->xUserData(pFts)!=0);
    memset(&debug, 0, sizeof(Fts5Buffer));

    int ip;
    double score = 0.0;

    if( bDebug ){
      fts5Bm25DebugContext(&rc, &debug, p);
      fts5Bm25DebugRow(&rc, &debug, p, pApi, pFts);
    }

    for(ip=0; rc==SQLITE_OK && ip<p->nPhrase; ip++){
      int iPrev = 0;
      int nHit = 0;
      int i = 0;
      i64 iPos = 0;

      while( rc==SQLITE_OK ){
        int bDone = pApi->xPoslist(pFts, ip, &i, &iPos);
        int iCol = FTS5_POS2COLUMN(iPos);

        if( (iCol!=iPrev || bDone) && nHit>0 ){
          int sz = 0;
          int idx = ip * p->nCol + iPrev;
          double bm25;
          rc = pApi->xColumnSize(pFts, iPrev, &sz);

          bm25 = (p->aIDF[idx] * nHit * (k1+1.0)) /
            (nHit + k1 * (1.0 - B + B * sz / p->aAvg[iPrev]));


          score = score + bm25;
          nHit = 0;
        }
        if( bDone ) break;
        nHit++;
        iPrev = iCol;
      }
    }

    if( rc==SQLITE_OK ){
      if( bDebug ){
        sqlite3_result_text(pCtx, (const char*)debug.p, -1, SQLITE_TRANSIENT);
      }else{
        sqlite3_result_double(pCtx, score);
      }
    }
    sqlite3_free(debug.p);
  }

  if( rc!=SQLITE_OK ){
    sqlite3_result_error_code(pCtx, rc);
  }
}

................................................................................
  struct Builtin {
    const char *zFunc;            /* Function name (nul-terminated) */
    void *pUserData;              /* User-data pointer */
    fts5_extension_function xFunc;/* Callback function */
    void (*xDestroy)(void*);      /* Destructor function */
  } aBuiltin [] = {
    { "bm25",      0, fts5Bm25Function,    0 },
    { "bm25debug", (void*)1, fts5Bm25Function,    0 },
    { "snippet",   0, fts5SnippetFunction, 0 },
    { "fts5_test", 0, fts5TestFunction,    0 },
  };

  int rc = SQLITE_OK;             /* Return code */
  int i;                          /* To iterate through builtin functions */

Changes to test/fts5ae.test.

225
226
227
228
229
230
231



232
233

































234
235
  } [list $res]
}

do_execsql_test 7.4 {
  SELECT fts5_test(t7, 'rowcount') FROM t7 WHERE t7 MATCH 'a';
} {5 5 5 5}







































finish_test








>
>
>
|

>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>


225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
  } [list $res]
}

do_execsql_test 7.4 {
  SELECT fts5_test(t7, 'rowcount') FROM t7 WHERE t7 MATCH 'a';
} {5 5 5 5}

#do_execsql_test 7.4 {
#  SELECT rowid, bm25debug(t7) FROM t7 WHERE t7 MATCH 'a';
#} {5 5 5 5}
#

#-------------------------------------------------------------------------
#
do_test 8.1 {
  execsql { CREATE VIRTUAL TABLE t8 USING fts5(x, y) }
  foreach {rowid x y} {
     0 {A o}   {o o o C o o o o o o o o}
     1 {o o B} {o o o C C o o o o o o o}
     2 {A o o} {o o o o D D o o o o o o}
     3 {o B}   {o o o o o D o o o o o o}
     4 {E o G} {H o o o o o o o o o o o}
     5 {F o G} {I o J o o o o o o o o o}
     6 {E o o} {H o J o o o o o o o o o}
     7 {o o o} {o o o o o o o o o o o o}
     9 {o o o} {o o o o o o o o o o o o}
  } {
    execsql { INSERT INTO t8(rowid, x, y) VALUES($rowid, $x, $y) }
  }
} {}

foreach {tn q res} {
  1 {a} {0 2}
  2 {b} {3 1}
  3 {c} {1 0}
  4 {d} {2 3}
  5 {g AND (e OR f)} {5 4}
  6 {j AND (h OR i)} {5 6}
} {
  do_execsql_test 8.2.$tn {
    SELECT rowid FROM t8 WHERE t8 MATCH $q ORDER BY bm25(t8) DESC;
  } $res
}


finish_test