allMus with unequal weights clashes with Dirichlet prior

[GuyBaele]

For example, the Ebola data set has 4 weights in total, with the weight for the intergenic region partition differing from the 3 weights of the codon positions. A deltaExchange combined with a Dirichlet prior (which we have been suggesting as of late) rejects all changes to ig.mu, given its different weight from the other positions (see screenshot).

The easiest way to fix this or make it more robust could be to simply use equal weights for all partitions, which in this case eliminates the problem:

[maxbiostat]

I thought we switched to the mu parametrisation exactly to escape DeltaExchange with unequal weights. What am I missing?

[rambaut]

Yes. Dirichlet can only be used if the delta exchange has equal weights. Which means it can't be used on unequal sized partitions. Hence the switch to _nu_s which have been adjust for the partition size and thus can use an equal delta exchange. But the logger rescales these back to _mu_s for easy interpretation.

[GuyBaele]

So I assume we keep things as they are for now, and eventually get rid of the 'return to classic priors' checkbox?

[rambaut]

Yes. But perhaps we should throw a parser error to help clarify this.

[GuyBaele]

Sounds like a plan.

[rambaut]

Got rid of classic prior option