Alex Herbert created STATISTICS-94:
--------------------------------------
Summary: Support median and quantile for long datatype
Key: STATISTICS-94
URL: https://issues.apache.org/jira/browse/STATISTICS-94
Project: Commons Statistics
Issue Type: Improvement
Reporter: Alex Herbert
Assignee: Alex Herbert
Add support to the {{Median}} and {{Quantile}} class for the {{long}} datatype.
Current support is for {{double[]}} and {{{}int[]{}}}. Via primitive conversion
these support float, short, char and byte if an array is appropriately
converted. However {{long}} cannot be supported by primitive conversion.
The current median and quantile result is returned as a {{{}double{}}}. This is
not appropriate for the result on a long array as the closest representable
double may be outside the range of the input data if the data have more than
53-bits in their integer representation. However if the integer value is
representable as less than 53-bits then the double will provide a fractional
part that cannot be returned if the result was a long. I suggest using the
{{StatisticResult}} for the return value allowing both double and long
representations:
{code:java}
StatisticResult r = Median.of(new long[] {Long.MIN_VALUE, Long.MAX_VALUE});
r.getAsDouble(); // -0.5
r.getAsLong(); // 0{code}
h2. Rounding
The current behaviour for {{StatisticResult}} is to return the integer
representations using the closest representation of the floating-point value
with ties towards positive infinity. So in the example above the median as a
long is 0.
This behaviour could be changed with a rounding mode parameter. Unfortunately
the current rounding behaviour is not represented by any mode in
[java.math.RoundingMode (Javadoc JDK
21)|https://docs.oracle.com/en/java/javase/21/docs/api/java.base/java/math/RoundingMode.html]
so this enum cannot be reused to set the mode. The rounding behaviour matches
[java.lang.Math.round(double) (Javadoc JDK
21)|https://docs.oracle.com/en/java/javase/21/docs/api/java.base/java/lang/Math.html#round(double)].
The closest match is RoundingMode.HALF_UP when above zero and
RoundingMode.HALF_DOWN when below zero. If rounding to nearest is only
concerned with the 0.5 boundary then support for rounding could add modes for
rounding ties towards: negative infinity; positive infinity: or zero
(floating-point truncation).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
