constrained_gaussian freezes

[sdwfrost]

The following code runs, but freezes - I suspect this is due to the choice of parameters, but no warning/error is thrown.

using EnsembleKalmanProcesses
using EnsembleKalmanProcesses.ParameterDistributions
constrained_gaussian("gamma", 0.5, 0.5, 0.0, 1.0)

[odunbar]

Thanks for pointing this out! Indeed, this is trying to create a bounded distribution within [0,1] but with mean of 0.5 and a standard deviation of 0.5. which hits the boundary.

For more outlandish values e.g. constrained_gaussian("gamma", 0.5, 0.8, 0.0, 1.0) it seems like our failsafe kicks in with an error.

But i guess for this value of sd (and some nearby), it tries to solve it without success.

Quick solves for you

• Using constrained_gaussian("gamma", 0.5, 0.4, 0.0, 1.0) Will put a lot of mass at the boundaries, (mu=1e-5, sig=4.06 this is even more extreme than e.g. the sig = 3.16 in the figure below)
• constrained_gaussian("gamma", 0.5, 0.3, 0.0, 1.0) Will create something close to uniform (mu=1e-5, sig=1.81 c.f. fig below)
  
  [from wikipedia logitnormal page]

https://en.wikipedia.org/wiki/Logit-normal_distribution

Actions

• I'll see if we can tighten up the error message tolerance
• in future we can improve the numerical solver robustness.

[odunbar]

Closing, now tolerances are sorted, and warning message implemented if solver will struggle.

[odunbar]

PR #266