Xiaochang Wu created SPARK-31505:
------------------------------------
Summary: Single precision floating point support in machine
learning
Key: SPARK-31505
URL: https://issues.apache.org/jira/browse/SPARK-31505
Project: Spark
Issue Type: Improvement
Components: ML
Affects Versions: 3.1.0
Reporter: Xiaochang Wu
MLlib assumed the algorithm data type to "double" in all its BLAS and algorithm
code. In some situations, there is no need to use high precision, end users may
want to trade precision with performance given machines have much more single
precision FP bandwiths than double precision ones.
It may require templating all FP data type and rewrite all existing code. We
would like to fast prototype some key algorithms and produce benchmarks on
using float instead of double. Feel free to comment on this if there is a need!
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
