Hierarquical structure over BSR missing block variables

[Víctor de Buen Remiro]

User should be able to define latent, prior and generic constraints over missing block variables.

A tipical situation is that an input or output is non stationary, so scalar truncated normal distributions are no too apropriate.

For example, if the original series requires a regular difference to be stationary, then user could write something like this

  knownValue[n-1] = missingResidual[n] + missingVariable[n] 

if n-th datum is unknown but (n-1)-th is known, or

  0 = missingResidual[n] + missingVariable[n] - missingVariable[n-1]

if both n-th and (n-1)-th data are unknown.