PQ seed is now also used for KMeansClustering.

vitrivr · Aug 14, 2024 · bc4d0aa · bc4d0aa
1 parent 1c75f83
commit bc4d0aa
Show file tree

Hide file tree

Showing 3 changed files with 4 additions and 5 deletions.
diff --git a/cottontaildb-dbms/src/main/kotlin/org/vitrivr/cottontail/dbms/index/pq/PQIndexConfig.kt b/cottontaildb-dbms/src/main/kotlin/org/vitrivr/cottontail/dbms/index/pq/PQIndexConfig.kt
@@ -12,14 +12,13 @@ import java.io.ByteArrayInputStream
 /**
  * Configuration class for [PQIndex].
  *
- * @author Gabriel Zihlmann & Ralph Gasser
+ * @author Gabriel Zihlmann
+ * @author Ralph Gasser
  * @version 1.3.0
  */
 data class PQIndexConfig(val distance: Name.FunctionName, val numCentroids: Int, val subspaces: Int, val seed: Int = System.currentTimeMillis().toInt()) : IndexConfig<PQIndex> {
 
     companion object {
-
-
         /** The maximum number of subspaces. We cap this at 32 to limit the code length.  */
         private const val MAXIMUM_NUMBER_OF_SUBSPACES = 32
 

diff --git a/...bms/src/main/kotlin/org/vitrivr/cottontail/dbms/index/pq/quantizer/MultiStageQuantizer.kt b/...bms/src/main/kotlin/org/vitrivr/cottontail/dbms/index/pq/quantizer/MultiStageQuantizer.kt
@@ -41,7 +41,7 @@ data class MultiStageQuantizer(val coarse: PQCodebook, val fine: Array<PQCodeboo
 
             /* Prepare k-means clusterer. */
             val reshape = distance.copy(dimensionsPerSubspace)
-            val random = SplittableRandom(System.currentTimeMillis())
+            val random = SplittableRandom(config.seed.toLong())
             val coarseClusterer = KMeansClusterer(config.numCoarseCentroids, distance.type, random)
             val fineClusterer = KMeansClusterer(config.numCentroids, distance.type, random)
 

diff --git a/...ms/src/main/kotlin/org/vitrivr/cottontail/dbms/index/pq/quantizer/SingleStageQuantizer.kt b/...ms/src/main/kotlin/org/vitrivr/cottontail/dbms/index/pq/quantizer/SingleStageQuantizer.kt
@@ -41,7 +41,7 @@ data class SingleStageQuantizer constructor(val codebooks: Array<PQCodebook>) {
 
             /* Prepare k-means clusterer. */
             val reshape = distance.copy(dimensionsPerSubspace)
-            val random = SplittableRandom(System.currentTimeMillis())
+            val random = SplittableRandom(config.seed.toLong())
             val clusterer = KMeansClusterer(config.numCentroids, reshape.type, random)
 
             /* Prepare codebooks. */