Motivation/Application
A log-normal distribution can be used with the current relative implementation by log-transforming the measurements and using only offset (which is practically the log of the scaling for non-transformed measurements). However, it's tedious as then the objective function is wrong: the log normal distribution has a measurement multiplicant in the log(2 pi sigma) which the normal does not have). Additionally, one has to transform all the values back for plotting etc...
Feature description Adapt the analytical formulas to support the log-normal distribution. This should not be a lot of changes, as the log-normal distribution is quite similar to the normal one. -- https://github.com/AMICI-dev/AMICI/blob/4d40911b777613bb9e1b09f26d84d9a82701ab13/python/sdist/amici/import_utils.py#L203-L207
Motivation/Application A log-normal distribution can be used with the current relative implementation by log-transforming the measurements and using only offset (which is practically the log of the scaling for non-transformed measurements). However, it's tedious as then the objective function is wrong: the log normal distribution has a measurement multiplicant in the
log(2 pi sigma)
which the normal does not have). Additionally, one has to transform all the values back for plotting etc...