[ https://issues.apache.org/jira/browse/MATH-1203?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14354525#comment-14354525 ]
Thomas Neidhart commented on MATH-1203: --------------------------------------- Sorry to bugger again, but there is something with the fix that should at least be documented. The getKernel method is protected, thus users may override it to return a different distribution, but the cumulativeProbability method now explicitly checks if the bin kernel is of type ConstantRealDistribution. Either we document this explicitly in the javadoc as a requirements when overriding the method, or we do a check on the variance of the kernel to decide if we have to do interpolation or not. > getKernel fails for buckets with only multiple instances of the same value in > random.EmpiricalDistribution > ---------------------------------------------------------------------------------------------------------- > > Key: MATH-1203 > URL: https://issues.apache.org/jira/browse/MATH-1203 > Project: Commons Math > Issue Type: Bug > Affects Versions: 3.4, 3.4.1 > Reporter: Kyle Kavanagh > > After loading a set of values into an EmpericalDistribution, assume that > there's a case where a single bin ONLY contains multiple instances of the > same value. In this case the standard deviation will equal zero. This will > fail when getKernel attempts to create a NormalDistribution. The other case > where stddev=0 is when there is only a single value in the bin, and this is > handled by returning a ConstantRealDistribution rather than a > NormalDistrbution. > See: https://issues.apache.org/jira/browse/MATH-984 -- This message was sent by Atlassian JIRA (v6.3.4#6332)