1. set initmode to Random (#655)

2. add initmode as an adjustable parameter to kmeans.conf and run.sh Signed-off-by: minmingz <minming.zhu@inte.com>

Author: minmingzhu

bin/functions/hibench_prop_env_mapping.py

@@ -94,6 +94,7 @@
  MAX_ITERATION="hibench.kmeans.max_iteration",
  K="hibench.kmeans.k",
  K_STORAGE_LEVEL="hibench.kmeans.storage.level",
+ K_INIT_MODE="hibench.kmeans.initializationmode",
  # For gmm
  INPUT_SAMPLE_GMM="hibench.gmm.input.sample",
  INPUT_CLUSTER_GMM="hibench.gmm.input.cluster",

bin/workloads/ml/kmeans/spark/run.sh

@@ -28,7 +28,7 @@ rmr_hdfs $OUTPUT_HDFS || true
 SIZE=`dir_size $INPUT_HDFS`
 START_TIME=`timestamp`
 
-run_spark_job com.intel.hibench.sparkbench.ml.DenseKMeans -k $K --numIterations $MAX_ITERATION --storageLevel $K_STORAGE_LEVEL $INPUT_HDFS/samples
+run_spark_job com.intel.hibench.sparkbench.ml.DenseKMeans -k $K --numIterations $MAX_ITERATION --storageLevel $K_STORAGE_LEVEL --initMode $K_INIT_MODE $INPUT_HDFS/samples
 END_TIME=`timestamp`
 
 gen_report ${START_TIME} ${END_TIME} ${SIZE}

conf/workloads/ml/kmeans.conf

@@ -49,3 +49,4 @@ hibench.workload.input	${hibench.hdfs.data.dir}/Kmeans/Input
 hibench.workload.output	${hibench.hdfs.data.dir}/Kmeans/Output
 
 hibench.kmeans.storage.level	MEMORY_ONLY
+hibench.kmeans.initializationmode Random

sparkbench/ml/src/main/scala/com/intel/sparkbench/ml/DenseKMeans.scala

@@ -39,7 +39,7 @@ object DenseKMeans {
  k: Int = -1,
  numIterations: Int = 10,
  storageLevel: String= "MEMORY_ONLY",
- initializationMode: InitializationMode = Parallel)
+ initializationMode: InitializationMode = Random)
 
  def main(args: Array[String]) {
  val defaultParams = Params()