[
https://issues.apache.org/jira/browse/PIG-2406?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13167296#comment-13167296
]
Scott Carey commented on PIG-2406:
----------------------------------
The above example of adding 2.51 many times and the result being slightly off
is due to the nature of floating point numbers, and is no different than what
happens in a database. Double precision math does that, there is no
'improvement' to make to it.
Ideally there would be another data type of higher or limitless precision than
double ('numeric' in some databases, BigDecimal in Java), but as Alan says,
these data types are extremely slow. They should be used judiciously.
For many use cases, a handful of functions for rounding down floating points is
all that is needed. Even rounding to one less decimal in the above example
gives the expected result.
In a database, I would have called a rounding function on the result. With
Postgres it would be something like 'round(sum(val), 6)' using the round
function here:
http://www.postgresql.org/docs/9.1/static/functions-math.html#FUNCTIONS-MATH-FUNC-TABLE
If Pig had more built-in numeric functions it would help many use cases where
exact precision is not required, but one may wish to print results for human
interpretation or compare two numbers after computation. To do these things
with double precision numbers you need to round numbers down or compare within
a tolerance rather than exact match. In Pig I have had to write code manually
to round numbers properly. For example, to round a floating point value to the
nearest integer you have to add 0.5 then cast to a long, or use a UDF.
> Improve double precision
> -------------------------
>
> Key: PIG-2406
> URL: https://issues.apache.org/jira/browse/PIG-2406
> Project: Pig
> Issue Type: New Feature
> Reporter: Anitha Raju
>
> Hi,
> When 2.51 (type: double) is added 24 times using SUM, it gives you result of
> 60.23999999999997.
> If a user wants to control the precision, that is, in the result he requires
> 60.24, it wouldnt be possible unless he implements a custom UDF.
> So if we can have a way to improve double precision,
> for eg: http://docs.oracle.com/javase/1.5.0/docs/api/java/math/BigDecimal.html
> Here with BigDecimal, one can improve the double precision and with
> MathContext, the user has control to set the precision.
> Can we have a new UDF which can do the same?
> Regards,
> Anitha
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
