tgvaughan / epiinf Goto Github PK

Hi Tim,
In Beauti, if the initial value of the removal probability under the tree prior is changed, but it is not also changed in the initial value option provided in the prior for the removal probability the default value ("1") is maintained in the final xml file.

Cheers,
carlo

This parameterization sometimes makes sense in terms of setting priors, but it can also improve mixing to operate on a variable that is bounded by [0,1] instead of one that is bounded by [0, infinity).

testing integration with slack

Hi Tim,

When I tried to implement EpiInf model based on my own sequences, I met such error:

"Start likelihood: -Infinity after 10 initialisation attempts
P(posterior) = -Infinity (was -Infinity)
P(prior) = -Infinity (was -Infinity)
P(EpiTreePrior.t:HK_BA_2_2_v1) = -Infinity (was -Infinity)
..."
Fatal exception: Could not find a proper state to initialise. Perhaps try another seed.

I've tried to adjust parameters in "Epi Tree Prior" panel. It doesn't work. But the example sequences is working. I'm just wondering if any special requirements for the sequencing data using EpiInf. Or any suggestions for this?

Many thanks in advance!

Cheers,
Ruopeng

While the existing tau leaping algorithm means that the computation time is no longer proportional to the total unsampled population size, it is still proportional to the number of nodes in the tree and hence the sampled population size. For datasets of 10^3 samples or more, this makes MCMC impractical.

The TL_prime branch contains work on a slightly different algorithm intended to leap over sampled as well as unsampled events, but the algorithm as it stands is broken.