Mean.evaluate() should use a two-pass algorithm
-----------------------------------------------
Key: MATH-174
URL: https://issues.apache.org/jira/browse/MATH-174
Project: Commons Math
Issue Type: Bug
Affects Versions: 1.1, 1.0
Reporter: Phil Steitz
Priority: Minor
Fix For: 1.2
Since it has access to the full array of stored data, Mean.evaluate(double[])
can improve its accuracy by executing a two-pass algorithm, first computing an
initial estimate using the definitional forumla and then correcting that value
by the mean deviation against that value. The attached patch makes the
correction and includes an algorithm reference. It also re-activates and
increases sensitivity in some of the certified data tests. The Michelson data
test fails with the current implementation, showing a difference in the 13th
significant digit.
This change will improve accuracy in DescriptiveStatistics.getMean and also in
Variance.evaluate() and DescriptiveStatistics.getVariance(). There is a cost
associated with the change - as it roughly doubles the arithmetic operations
required to compute the mean. Since it only applies to the "stored array"
implementation and the implementation will be pluggable in 1.2, my preference
is to move ahead with this change.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
