For ML applications, the best setting to set the number of partitions
to match the number of cores to reduce shuffle size. You have 3072
partitions but 128 executors, which causes the overhead. For the
MultivariateOnlineSummarizer, we plan to add flags to specify what
need to be computed to reduce
I'm not super familiar with this part of the code, but from taking a quick
look:
a) the code creates a MultivariateOnlineSummarizer, which stores 7 doubles
per feature (mean, max, min, etc. etc.)
b) The limit is on the result size from *all* tasks, not from one task.
You start with 3072 tasks
c)
Hi,
I am trying to validate our modeling data pipeline by running
LogisticRegressionWithLBFGS on a dataset with ~3.7 million features,
basically to compute AUC. This is on Spark 1.3.0.
I am using 128 executors with 4 GB each + driver with 8 GB. The number of
data partitions is 3072
The