set tolerance to 0 (Intel-bigdata#651)

* 1. set tol to 0 Signed-off-by: minmingz <[email protected]> * 1. remove convergedist Signed-off-by: minmingz <[email protected]> * 1. add code comment Signed-off-by: minmingz <[email protected]> Co-authored-by: minmingz <[email protected]>
henrque · Dec 2, 2020 · f55862a · f55862a
1 parent fe3738e
commit f55862a
Show file tree

Hide file tree

Showing 2 changed files with 1 addition and 7 deletions.
diff --git a/conf/workloads/ml/kmeans.conf b/conf/workloads/ml/kmeans.conf
@@ -4,50 +4,43 @@ hibench.kmeans.tiny.num_of_samples		30000
 hibench.kmeans.tiny.samples_per_inputfile	6000
 hibench.kmeans.tiny.max_iteration		5
 hibench.kmeans.tiny.k				10
-hibench.kmeans.tiny.convergedist		0.5
 hibench.kmeans.small.num_of_clusters		5
 hibench.kmeans.small.dimensions			20
 hibench.kmeans.small.num_of_samples		3000000
 hibench.kmeans.small.samples_per_inputfile	600000
 hibench.kmeans.small.max_iteration		5
 hibench.kmeans.small.k				10
-hibench.kmeans.small.convergedist		0.5
 hibench.kmeans.large.num_of_clusters		5
 hibench.kmeans.large.dimensions			20
 hibench.kmeans.large.num_of_samples		20000000
 hibench.kmeans.large.samples_per_inputfile	4000000
 hibench.kmeans.large.max_iteration		5
 hibench.kmeans.large.k				10
-hibench.kmeans.large.convergedist		0.5
 hibench.kmeans.huge.num_of_clusters		5
 hibench.kmeans.huge.dimensions			20
 hibench.kmeans.huge.num_of_samples		100000000
 hibench.kmeans.huge.samples_per_inputfile	20000000
 hibench.kmeans.huge.max_iteration		5
 hibench.kmeans.huge.k				10
-hibench.kmeans.huge.convergedist		0.5
 hibench.kmeans.gigantic.num_of_clusters		5
 hibench.kmeans.gigantic.dimensions		20
 hibench.kmeans.gigantic.num_of_samples		200000000
 hibench.kmeans.gigantic.samples_per_inputfile	40000000
 hibench.kmeans.gigantic.max_iteration		5
 hibench.kmeans.gigantic.k			10
-hibench.kmeans.gigantic.convergedist		0.5
 hibench.kmeans.bigdata.num_of_clusters		5
 hibench.kmeans.bigdata.dimensions		20
 hibench.kmeans.bigdata.num_of_samples		1200000000
 hibench.kmeans.bigdata.samples_per_inputfile	40000000
 hibench.kmeans.bigdata.max_iteration		10
 hibench.kmeans.bigdata.k			10
-hibench.kmeans.bigdata.convergedist		0.5
 
 hibench.kmeans.num_of_clusters		${hibench.kmeans.${hibench.scale.profile}.num_of_clusters}
 hibench.kmeans.dimensions		${hibench.kmeans.${hibench.scale.profile}.dimensions}
 hibench.kmeans.num_of_samples		${hibench.kmeans.${hibench.scale.profile}.num_of_samples}
 hibench.kmeans.samples_per_inputfile	${hibench.kmeans.${hibench.scale.profile}.samples_per_inputfile}
 hibench.kmeans.max_iteration		${hibench.kmeans.${hibench.scale.profile}.max_iteration}
 hibench.kmeans.k			${hibench.kmeans.${hibench.scale.profile}.k}
-hibench.kmeans.convergedist		${hibench.kmeans.${hibench.scale.profile}.convergedist}
 hibench.kmeans.base.hdfs		${hibench.hdfs.data.dir}/Kmeans
 hibench.kmeans.input.sample		${hibench.workload.input}/samples
 hibench.kmeans.input.cluster		${hibench.workload.input}/cluster

diff --git a/sparkbench/ml/src/main/scala/com/intel/sparkbench/ml/DenseKMeans.scala b/sparkbench/ml/src/main/scala/com/intel/sparkbench/ml/DenseKMeans.scala
@@ -110,6 +110,7 @@ object DenseKMeans {
       .setK(params.k)
       .setMaxIter(params.numIterations)
       .setSeed(1L)
+      .setTol(0)       //set convergence to 0, aiming to execute the number of iterations of the algorithm without being affected by the convergence parameters.
       .fit(examples)
 
     val cost = model.summary.trainingCost