commons-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Phil Steitz (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (MATH-1034) Add binomial test
Date Mon, 30 Sep 2013 21:14:26 GMT

    [ https://issues.apache.org/jira/browse/MATH-1034?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13782239#comment-13782239
] 

Phil Steitz edited comment on MATH-1034 at 9/30/13 9:13 PM:
------------------------------------------------------------

Initial commit in r1527777 with the following changes:
* Eliminated references to R equivalents in AlternativeHypothesis.  This is useful but commits
us to keeping up with R changes, so best to drop.
* Minor edits to javadoc
* Replaced license header with standard header (same content, just format)
* Added $Id$ and @version tags
* Made distribution final in implementation code
* Fixed check style complaints (mostly {} around if blocks)
* Externalized exception strings
* Used NullArgumentException for null alternativeHypothesis
* Used MathInternalError for bad enum value (should be impossible)
* Added verification of null AlternativeHypothesis behavior in unit test

Still left to do:
* Improve precision sensitivity of unit tests.  Use R options(digits=12) if necessary to display
more digits of accuracy.  Mention the version of R used to generate the reference values in
test comments.
* Add more unit tests.
* Consider the following changes / improvements:
0) To be consistent with the other inference tests, specify "significance level" rather than
"confidence level" in test parameters (so e.g .95 becomes .05)
1) I suspect the efficiency of the code for the two-sided p-value computation can be improved.
 Instead of effectively searching for the numberOfSuccesses as the code does now, start by
comparing the number of successes to the mean = numberOfTrials * p.  If the number of successes
is greater than the mean, return twice the tail probability above this value; if it is less,
return twice the tail probability below it.




was (Author: psteitz):
Initial commit in r1527777 with the following changes:
* Eliminated references to R equivalents in AlternativeHypothesis.  This is useful but commits
us to keeping up with R changes, so best to drop.
* Minor edits to javadoc
* Replaced license header with standard header (same content, just format)
* Added $Id$ and @version tags
* Made distribution final in implementation code
* Fixed check style complaints (mostly {} around if blocks)
* Externalized exception strings
* Used NullArgumentException for null alternativeHypothesis
* Used MathInternalError for bad enum value (should be impossible)
* Added verification of null AlternativeHypothesis behavior in unit test

Still left to do:
* Improve precision sensitivity of unit tests.  Use R options(digits=12) if necessary to display
more digits of accuracy.  Mention the version of R used to generate the reference values in
test comments.
* Add more unit tests.
* Consider the following changes / improvements:
0) To be consistent with the other inference tests, specify "significance level" rather than
"confidence level" in test parameters (so e.g .95 becomes .05)
1) I suspect the efficiency of the code for the tee-sided p-value computation can be improved.
 Instead of effectively searching for the numberOfSuccesses as the code does now, start by
comparing the number of successes to the mean = numberOfTrials * p.  If the number of successes
is greater than the mean, return twice the tail probability above this value; if it is less,
return twice the tail probability below it.



> Add binomial test
> -----------------
>
>                 Key: MATH-1034
>                 URL: https://issues.apache.org/jira/browse/MATH-1034
>             Project: Commons Math
>          Issue Type: Wish
>    Affects Versions: 3.2
>            Reporter: Thorsten Schäfer
>            Priority: Minor
>         Attachments: binomialTest.patch, binomialTest.patch
>
>
> A binomial test would be a nice addition to commons-math. I might supply a patch in the
near future. I guess the interface should be similar to the other tests, i.e., a method to
get the p-value and a method returning a boolean indicating reject/non-reject.
> Is there a policy about using Enumerations in commons-math? For instance, in R you can
test two-sided, less or greater. This could be done using an enumeration in Java, but I'm
not sure if this is discouraged for backward compatibility reasons...



--
This message was sent by Atlassian JIRA
(v6.1#6144)

Mime
View raw message