Feature requests for multi.ace about disprity HOT 2 CLOSED

Hi Armin,

That’s a good suggestion! Thanks for asking:

• 1. Could you incorporate the log-likelihood and the AIC value in the multi.ace() output?

I’ve update multi.ace in dispRity 1.5.9 to handle the extra argument estimation.details to extract some specific output from castor::asr_mk_model. For example, you can now also get the likelihood from the each model (per character and per tree) and their transition matrices using multi.ace(…, estimation.details = c("loglikelihood", "transition_matrix")) (you can also get the ancestral states specific likelihood values by asking for "ancestral_likelihoods" for example… Let me know if that works for you. Unfortunately for the AIC, you’ll have to compute yourself I’m afraid (but see point 2)

• 2. This one might be a bit more work: Could you implement a summary function for the ancestral estimates (i. e. summarize ancestral states across a set of trees)? Maybe the mean of the ancestral estimates across the analyses set of trees (for a fixed topology it is relatively straightforward, but for trees that differ in topology it might be a bit more work)? I have also attached a little function, that is meant to produce the mean of ancestral state estimates based on the ace() output (unfortunately I only saw your multi.ace() function later on, but I reckon this could be easily modified accordingly).

I think this would be really useful (for example for calculating AIC values ;)) and this could be incorporated to some specific print, summary and plot functions however, this would require some more time and some more thinking: although it’ll be possible to get some average trait likelihood values and stuff like that, I’m not sure yet how to summarise variance and “average” with discrete trait values. But maybe by summarising the variance per character and per tree that would work? Like character table where each cell contains either the modal character, the average scaled likelihood for that character across the multiple trees and the variance in scaled likelihood across trees. Something like that:

Modal character:

Scaled likelihood for modal character:

Scaled likelihood variance:

This could easily show that character 1 and 2 for taxa 1 are pretty well estimated (i.e. with confidence) whereas there are a bit shittier for taxa 2 (with character 1 for taxa 2 maybe being better because of uncertainty, etc...).

Would that work? Unfortunately I won't have time to do it these days but I'll put it on my TODO list for a future version. Alternatively, I encourage you to propose me an implementation if you have time!

Cheers,
Thomas

from disprity.

Hi Thomas,

cheers, thanks for the update!
Yes, what you propose looks like the way to go - I will see, whether I can find the time to implement your suggestions. For now, I am also happy with my primitive solution, which just calculates the mean likelihood values for each ancestral state value.

Thanks a lot for your help!

from disprity.