You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "zhengruifeng (JIRA)" <ji...@apache.org> on 2019/07/24 10:21:00 UTC
[jira] [Updated] (SPARK-28499) Optimize MinMaxScaler

     [ https://issues.apache.org/jira/browse/SPARK-28499?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

zhengruifeng updated SPARK-28499:
---------------------------------
    Description: 
current impl of MinMaxScaler has some small places to be optimized:

1, avoid call param getter in udf.

If I remember correctly, there was some tickets and prs about this, calling param getter in udf or map function, will significantly slow down the computation.

2, for a constant dim, the transformed value is also a constant value, which can be precomputed.

3, for a usual dim (i-th), the value is update by

values(i) = (values(i) - minArray(i)) / range(i) * scale + $(min)

here, we can precompute  scale / range, so that a division can be skipped.

  was:
current impl of MinMaxScaler has some small places to be optimized:

1, avoid call param getter in udf.

If I remember correctly, there was some tickets and prs about this, calling param getter in udf or map function, will significantly slow down the computation.

2, for a constant dim, the transformed value is also a constant value, which can be precomputed.

3, for a usual dim (i-th), the value is update by

values(i) = (values(i) - minArray(i)) / range(i) * scale + $(min)

here, we can precompute range * scale, so that a division can be skipped.


> Optimize MinMaxScaler
> ---------------------
>
>                 Key: SPARK-28499
>                 URL: https://issues.apache.org/jira/browse/SPARK-28499
>             Project: Spark
>          Issue Type: Improvement
>          Components: ML
>    Affects Versions: 3.0.0
>            Reporter: zhengruifeng
>            Priority: Minor
>
> current impl of MinMaxScaler has some small places to be optimized:
> 1, avoid call param getter in udf.
> If I remember correctly, there was some tickets and prs about this, calling param getter in udf or map function, will significantly slow down the computation.
> 2, for a constant dim, the transformed value is also a constant value, which can be precomputed.
> 3, for a usual dim (i-th), the value is update by
> values(i) = (values(i) - minArray(i)) / range(i) * scale + $(min)
> here, we can precompute  scale / range, so that a division can be skipped.



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org