Modifications to overcome a few issues

Hi Nicola,

I have forked the BASTA repository to make some changes to fix a couple issues I have encountered. If it's okay with you, it would be great if these modifications can be incorporated into the master branch. The issues are

(1) The prior on popSize with dimension > 1 but uniform values (Issue #3, in the issue database)

(2) Numerical instability encounter when sampling ancestral states with unconvincing parameter values of the structured coalescent model.

The details of (1) have been described in the issue database.

The ancestral state sampling in a BASTA framework is not part of the MCMC, as the structured coalescent integrates out the migration history. Therefore the sampling of the ancestral states can be done during the MCMC or after the MCMC.

The problem with doing it during the MCMC prior to the accept/reject step is that if the parameter values unconvincing, in other words, producing -Inf coalescent likelihood, you would encounter numerical instability when sampling of the ancestral states. The numerical instability could cause the MCMC to terminate even though ancestral states do not play a part in the MCMC. Consequently, for design and practical reasons, it would be best to do the ancestral state sampling after the completion of MCMC.

I have created additional classes and modified some of the existing classes to handle this. However, I also retain the function of doing ancestral-state-sampling during the MCMC (as it has been in BASTA). This means that you have to option of doing ancestral-state-sampling during OR after the MCMC.

Thanks,

Jessie