Changeset: 08cf5e383bd5 for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=08cf5e383bd5
Modified Files:
monetdb5/extras/rdf/rdfschema.c
monetdb5/extras/rdf/rdfschema.h
Branch: rdf
Log Message:
Filter for FK relationships
diffs (183 lines):
diff --git a/monetdb5/extras/rdf/rdfschema.c b/monetdb5/extras/rdf/rdfschema.c
--- a/monetdb5/extras/rdf/rdfschema.c
+++ b/monetdb5/extras/rdf/rdfschema.c
@@ -2594,6 +2594,7 @@ void mergeCSbyS4(CSset *freqCSset, CSlab
mergecs1 = (CS*)&(freqCSset->items[tmpParentIdx]);
mergecs2 = (CS*)&(freqCSset->items[freqId1]);
+ printf("MaxCS: Merge freqCS %d and freqCS %d \n",
tmpParentIdx, freqId1);
mergeConsistsOf(mergecs1, mergecs2);
}
@@ -3125,7 +3126,7 @@ void doMerge(CSset *freqCSset, int ruleN
addCStoSet(freqCSset,*mergecs);
updateLabel(ruleNum, freqCSset, labels, 1,
freqCSset->numCSadded - 1, freqId1, freqId2, name, ontmetadata,
ontmetadataCount, NULL, -1);
free(mergecs);
-
+
mergecsId[0]++;
}
else if (cs1->parentFreqIdx == -1 && cs2->parentFreqIdx != -1){
@@ -3216,29 +3217,44 @@ str mergeMaxFreqCSByS1(CSset *freqCSset,
tmpCount = 0;
for (k = 0; k < labelStat->lstCount[i]; k++){
freqId1 = labelStat->freqIdList[i][k];
- if ((*labels)[freqId1].isOntology == 1) break;
+ if ((*labels)[freqId1].isOntology == 1) {
+ cs1 = &(freqCSset->items[freqId1]);
+ #if NOT_MERGE_DIMENSIONCS
+ if (cs1->type == DIMENSIONCS) continue;
+ #endif
+ tmpCount++;
+ break;
+ }
}
- cs1 = &(freqCSset->items[freqId1]);
for (j = k+1; j < labelStat->lstCount[i]; j++){
freqId2 = labelStat->freqIdList[i][j];
cs2 = &(freqCSset->items[freqId2]);
#if NOT_MERGE_DIMENSIONCS
- if (cs2->type == DIMENSIONCS) continue;
+ if (cs2->type == DIMENSIONCS)
+ continue;
#endif
if ((*labels)[freqId2].isOntology == 1){
+ printf("Merge FreqCS %d and FreqCS %d
by Ontology name \n", freqId1, freqId2);
doMerge(freqCSset, S1, cs1, cs2,
freqId1, freqId2, mergecsId, labels, ontmetadata, ontmetadataCount, *name);
+ printf("Number of added cs in freqCS:
%d \n", freqCSset->numCSadded);
tmpCount++;
}
}
- fprintf(fout, " %d names are same by Ontology. MergedCS
has %d prop. \n", tmpCount, freqCSset->items[freqCSset->numCSadded -1].numProp);
+ fprintf(fout, " %d freqCS merged as having same name by
Ontology. MergedCS has %d prop. \n", tmpCount,
freqCSset->items[freqCSset->numCSadded -1].numProp);
//For Type
tmpCount = 0;
for (k = 0; k < labelStat->lstCount[i]; k++){
freqId1 = labelStat->freqIdList[i][k];
- if ((*labels)[freqId1].isType == 1) break;
+ if ((*labels)[freqId1].isType == 1) {
+ cs1 = &(freqCSset->items[freqId1]);
+ #if NOT_MERGE_DIMENSIONCS
+ if (cs1->type == DIMENSIONCS) continue;
+ #endif
+ tmpCount++;
+ break;
+ }
}
- cs1 = &(freqCSset->items[freqId1]);
for (j = k+1; j < labelStat->lstCount[i]; j++){
freqId2 = labelStat->freqIdList[i][j];
cs2 = &(freqCSset->items[freqId2]);
@@ -3246,19 +3262,27 @@ str mergeMaxFreqCSByS1(CSset *freqCSset,
if (cs2->type == DIMENSIONCS) continue;
#endif
if ((*labels)[freqId2].isType == 1){
+ printf("Merge FreqCS %d and FreqCS %d
by Type name \n", freqId1, freqId2);
doMerge(freqCSset, S1, cs1, cs2,
freqId1, freqId2, mergecsId, labels, ontmetadata, ontmetadataCount, *name);
+ printf("Number of added cs in freqCS:
%d \n", freqCSset->numCSadded);
tmpCount++;
}
}
- fprintf(fout, " %d names are same by TYPE. MergedCS has
%d prop. \n", tmpCount, freqCSset->items[freqCSset->numCSadded -1].numProp);
+ fprintf(fout, " %d freqCS merged as having same name by
TYPE. MergedCS has %d prop. \n", tmpCount,
freqCSset->items[freqCSset->numCSadded -1].numProp);
//For FK
tmpCount = 0;
for (k = 0; k < labelStat->lstCount[i]; k++){
freqId1 = labelStat->freqIdList[i][k];
- if ((*labels)[freqId1].isFK == 1) break;
+ if ((*labels)[freqId1].isFK == 1) {
+ cs1 = &(freqCSset->items[freqId1]);
+ #if NOT_MERGE_DIMENSIONCS
+ if (cs1->type == DIMENSIONCS) continue;
+ #endif
+ tmpCount++;
+ break;
+ }
}
- cs1 = &(freqCSset->items[freqId1]);
for (j = k+1; j < labelStat->lstCount[i]; j++){
freqId2 = labelStat->freqIdList[i][j];
cs2 = &(freqCSset->items[freqId2]);
@@ -3266,13 +3290,14 @@ str mergeMaxFreqCSByS1(CSset *freqCSset,
if (cs2->type == DIMENSIONCS) continue;
#endif
if ((*labels)[freqId2].isFK == 1){
+ printf("Merge FreqCS %d and FreqCS %d
by FK name \n", freqId1, freqId2);
doMerge(freqCSset, S1, cs1, cs2,
freqId1, freqId2, mergecsId, labels, ontmetadata, ontmetadataCount, *name);
+ printf("Number of added cs in freqCS:
%d \n", freqCSset->numCSadded);
tmpCount++;
}
}
#endif /* USE_MULTIWAY_MERGING */
- fprintf(fout, " %d names are same by FK. MergedCS has
%d prop. \n", tmpCount, freqCSset->items[freqCSset->numCSadded -1].numProp);
-
+ fprintf(fout, " %d freqCS merged as having same name by
FK. MergedCS has %d prop. \n", tmpCount, freqCSset->items[freqCSset->numCSadded
-1].numProp);
#if OUTPUT_FREQID_PER_LABEL
@@ -3577,7 +3602,8 @@ void mergeCSByS3S5(CSset *freqCSset, CSl
if (simscore > SIM_TFIDF_THRESHOLD){
#else
if (simscore > SIM_THRESHOLD) {
- #endif
+ #endif
+ //printf("S3S5: merge freqCS %d and freqCS %d
\n", freqId1, freqId2);
//Check whether these CS's belong to any mergeCS
if (cs1->parentFreqIdx == -1 &&
cs2->parentFreqIdx == -1){ /* New merge */
mergecs = mergeTwoCSs(*cs1,*cs2,
freqId1,freqId2, *mergecsId);
@@ -5087,7 +5113,7 @@ CSrel* getFKBetweenTableSet(CSrel *csrel
refinedCsRel = initCSrelset(numTables);
for (i = 0; i < numRel; ++i) {
- if (csrelFreqSet[i].numRef == 0 || freqCSset->items[i].coverage
> MINIMUM_TABLE_SIZE) continue; // ignore CS without relations
+ if (csrelFreqSet[i].numRef == 0) continue; // ignore CS without
relations
assert(freqCSset->items[i].parentFreqIdx == -1);
rel = csrelFreqSet[i];
from = mfreqIdxTblIdxMapping[i];
@@ -5112,7 +5138,15 @@ CSrel* getFKBetweenTableSet(CSrel *csrel
}
assert(propIdx < freqCSset->items[i].numProp);
- if (csPropTypes[from].lstPropTypes[propIdx].propCover *
MIN_FK_PROPCOVERAGE > rel.lstCnt[j]) continue;
+
+ //Filtering: For big size table, if large number of
prop's instances need to refer to a certain table
+ // else, all instances of that prop must refer to the
certain table
+ if (freqCSset->items[i].coverage > MINIMUM_TABLE_SIZE){
+ if
(csPropTypes[from].lstPropTypes[propIdx].propCover * MIN_FK_PROPCOVERAGE >
rel.lstCnt[j]) continue;
+ }
+ else{
+ if
(csPropTypes[from].lstPropTypes[propIdx].propCover != rel.lstCnt[j]) continue;
+ }
assert(to < numTables);
addReltoCSRelWithFreq(from, to, rel.lstPropId[j],
rel.lstCnt[j], rel.lstBlankCnt[j], &refinedCsRel[from]);
@@ -5479,6 +5513,7 @@ RDFextractCSwithTypes(int *ret, bat *sba
curT = clock();
printf("Merging with S1 took %f. (Number of mergeCS: %d | NumconsistOf:
%d) \n", ((float)(curT - tmpLastT))/CLOCKS_PER_SEC, curNumMergeCS,
countNumberConsistOfCS(freqCSset));
+ printf("Number of added CS after S1: %d \n", freqCSset->numCSadded);
tmpLastT = curT;
/* ---------- S4 ------- */
diff --git a/monetdb5/extras/rdf/rdfschema.h b/monetdb5/extras/rdf/rdfschema.h
--- a/monetdb5/extras/rdf/rdfschema.h
+++ b/monetdb5/extras/rdf/rdfschema.h
@@ -129,7 +129,7 @@ typedef struct PropStat {
IR_DIMENSION_THRESHOLD_PERCENTAGE * totalFrequency
Number of IR
references should be several times larger than the CS frequency
*/
-#define NOT_MERGE_DIMENSIONCS 1
+#define NOT_MERGE_DIMENSIONCS 1 /* Default: 1, 0: Is for
example data */
#define FILTER_INFREQ_FK_FOR_IR 1 /* We filter out all
the dirty references from a CS */
#define FILTER_THRESHOLD_FK_FOR_IR 0.1 /* The FK that their frequency
< FILTER_THRESHOLD_FK_FOR_IR * FreqCS's frequency */
@@ -204,6 +204,7 @@ typedef struct SubCSSet{
#define MIN_PERCETAGE_S6 5 // Merge all CS refered by more than
1/MIN_PERCETAGE_S6 percent of a CS via one property
#define MIN_FROMTABLE_SIZE_S6 100 // The minimum size of the "from" table in
S6. Meaning that
// the CS's to-be-merged in this rule must
cover > MIN_FROMTABLE_SIZE_S6 / MIN_PERCETAGE_S6 triples
+//#define MIN_FROMTABLE_SIZE_S6 1 /* For example data */
#define MINIMUM_TABLE_SIZE 10000 //The minimum number of triples coverred by
a table (i.e., a final CS)
#define SAMPLE_FILTER_THRESHOLD 1 // SAMPLE_FILTER_THRESHOLD/ 100
#define HIGH_REFER_THRESHOLD 5
_______________________________________________
checkin-list mailing list
checkin-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/checkin-list
