Gaussian & LogNormal distributions

[dhardy]

Negative standard deviation

Currently Normal::new checks that std_dev is not negative. I see no need to keep this check (GSL lacks it), though it should be pointed out that effectively the standard deviation will be std_dev.abs().

The only drawback is that calculations erroneously producing a negative value would no longer be detected, though I doubt this often happens (and non-negative erroneous values would still not be detected).

[dhardy]

LogNormals by mean and CV

For another project, we decided the most convenient way to parameterise log-normals was by the (linear) mean and coefficient of variation. Code found here, written by myself.

We should add this functionality, and clarify the parameters to LogNormal::new in the doc.

[dhardy]

Z-scores and correlated samples

Sometimes it can be useful to save a z-score (sample from a standard normal), and use it to sample from a parameterised Normal or LogNormal distribution, allowing correlated samples or a disconnect between the RNG machinery and parmeterisation.

Additionally, it is possible to generate partially-correlated z-scores (with correlation measured in linear space for Normal samples, or in log-space for LogNormal samples).

Example code for this here (also written by myself).

[vks]

We should add this functionality, and clarify the parameters to LogNormal::new in the doc.

I tried to clarify the parameter names in 466a0fc.

[vks]

@dhardy Can this be closed?

[dhardy]

Yes, I think so.