liuyu/pgvector
1
0
Fork 0
mirror of https://github.com/pgvector/pgvector.git synced 2026-06-30 17:51:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

Improved k-means types [skip ci]

Browse Source
This commit is contained in:
Andrew Kane
2024-04-11 20:00:07 -07:00
parent 8d9400bae3
commit cca0edf458
1 changed files with 12 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
src/ivfkmeans.c
View File

@@ -297,7 +297,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, IvfflatTyp
newcdist = palloc(newcdistSize);
aggCenters = VectorArrayInit(numCenters, dimensions, VECTOR_SIZE(dimensions));
for (int64 j = 0; j < numCenters; j++)
for (int j = 0; j < numCenters; j++)
{
Vector *vec = (Vector *) VectorArrayGet(aggCenters, j);
@@ -460,11 +460,11 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, IvfflatTyp
}
/* Step 4: For each center c, let m(c) be mean of all points assigned */
for (int64 j = 0; j < numCenters; j++)
for (int j = 0; j < numCenters; j++)
{
Vector *vec = (Vector *) VectorArrayGet(aggCenters, j);
for (int64 k = 0; k < dimensions; k++)
for (int k = 0; k < dimensions; k++)
vec->x[k] = 0.0;
centerCounts[j] = 0;
@@ -480,14 +480,14 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, IvfflatTyp
{
Vector *vec = (Vector *) VectorArrayGet(samples, j);
for (int64 k = 0; k < dimensions; k++)
for (int k = 0; k < dimensions; k++)
aggCenter->x[k] += vec->x[k];
}
else if (type == IVFFLAT_TYPE_HALFVEC)
{
HalfVector *vec = (HalfVector *) VectorArrayGet(samples, j);
for (int64 k = 0; k < dimensions; k++)
for (int k = 0; k < dimensions; k++)
aggCenter->x[k] += HalfToFloat4(vec->x[k]);
}
else
@@ -496,7 +496,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, IvfflatTyp
centerCounts[closestCenter] += 1;
}
for (int64 j = 0; j < numCenters; j++)
for (int j = 0; j < numCenters; j++)
{
Vector *vec = (Vector *) VectorArrayGet(aggCenters, j);
@@ -504,19 +504,19 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, IvfflatTyp
{
/* Double avoids overflow, but requires more memory */
/* TODO Update bounds */
for (int64 k = 0; k < dimensions; k++)
for (int k = 0; k < dimensions; k++)
{
if (isinf(vec->x[k]))
vec->x[k] = vec->x[k] > 0 ? FLT_MAX : -FLT_MAX;
}
for (int64 k = 0; k < dimensions; k++)
for (int k = 0; k < dimensions; k++)
vec->x[k] /= centerCounts[j];
}
else
{
/* TODO Handle empty centers properly */
for (int64 k = 0; k < dimensions; k++)
for (int k = 0; k < dimensions; k++)
vec->x[k] = RandomDouble();
}
}
@@ -544,7 +544,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, IvfflatTyp
}
/* Step 5 */
for (int64 j = 0; j < numCenters; j++)
for (int j = 0; j < numCenters; j++)
newcdist[j] = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(VectorArrayGet(centers, j)), PointerGetDatum(VectorArrayGet(newCenters, j))));
for (int64 j = 0; j < numSamples; j++)
@@ -562,11 +562,11 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, IvfflatTyp
/* Step 6 */
/* We reset r(x) before Step 3 in the next iteration */
for (int64 j = 0; j < numSamples; j++)
for (int j = 0; j < numSamples; j++)
upperBound[j] += newcdist[closestCenters[j]];
/* Step 7 */
for (int64 j = 0; j < numCenters; j++)
for (int j = 0; j < numCenters; j++)
VectorArraySet(centers, j, VectorArrayGet(newCenters, j));
if (changes == 0 && iteration != 0)