Alex Herbert created RNG-166:
--------------------------------
Summary: LogNormalSampler should allow negative scale (mean of the
natural logarithm of the distribution values)
Key: RNG-166
URL: https://issues.apache.org/jira/browse/RNG-166
Project: Commons RNG
Issue Type: Bug
Components: sampling
Affects Versions: 1.4
Reporter: Alex Herbert
The [Log-normal
distribution|https://en.wikipedia.org/wiki/Log-normal_distribution] is defined
with respect to the mean and standard deviation of the natural logarithm of the
distribution values. The mean can be any number in (-inf, inf).
Currently the LogNormalSampler only allows the mean to be in [0, inf).
The sampler also refers to the parameters as shape and scale. They should be
updated to:
* shape -> mu : Mean of the natural logarithm of the distribution values.
* scale -> sigma : Standard deviation of the natural logarithm of the
distribution values.
This bug was discovered when testing the LogNormalDistribution in Commons
Statistics.
Note the original distribution in Commons Math 3 allowed a negative mean. The
parameters in CM3 used the terms shape and scale. These were ported to Commons
RNG but have been corrected to mu and sigma in Commons Statistics.
I have verified that if the condition for a positive mean is dropped from the
LogNormalSampler then the LogNormalDistribution can create a sampler and
satisfy a chi-squared test on the samples.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
