dmcglinn
commented
2 years ago Hey @T-Engel I'm picking this task back up. Any thoughts here?
Also I have some questions about two edge cases for the function invChat.
Specifically, when the sample size of the target coverage is calculated using invChat an error will result when there are no singletons in the community. See this line:
https://github.com/MoBiodiv/mobr/blob/dev/R/beta_C.R#L98
If f1 == 0 then mm will become mathematically undefined.
The Chao extrapolation method requires that there are singletons otherwise you are assumed to be at the asymptote and coverage is 1. So in this situation in which f1 == 0 I think we should just return the number of individuals in the sample.
From that same line of code you can see that mm will also be undefined if C == 1 (coverage is 1). Any suggestions for a better value of mm in this case?
Thanks!
T-Engel
sablowes
I'm trying to shore up the calculation of beta diversity across the various metrics we have proposed. In an effort to keep methods consistent across metrics and to keep practitioners from dwelling on beta diversity in isolation I was thinking of sunsetting the
beta_Cfunction and instead changing it to the functioncalc_S_Cwhich will calculate richness for a given coverage level or if no coverage is supplied it will find an ideal coverage and then compute S. This function will then be used in the same standard flow other metrics undergo for calculating beta-diversity.To calculate
beta_Cwhat I'm proposing a user would do is:calc_comm_div(my_comm, index = 'S_C', scales = 'beta')that function will make a call to:calc_divwhich will make two calls (gamma and alpha) to:calc_S_Cwhich will ultimately userarefactionto compute the S_n values at the appropriate coverage.The default behavior of calc_comm_div will be:
calc_comm_div(my_comm, index = 'S_C')which would returnS_Cat the alpha and gamma scales andbeta_S_Cas the ratio.The
calc_comm_divfunction will also be able to call the confidence interval functions that I've been working on so it's the wrapper that pulls it all together for all functions.My questions for folks (pinging @T-Engel) are:
1) does that sound reasonable? 2) are you ok with
calc_S_CI thought it was slightly better thancalc_SCorcalc_Sc3) is the target coverage function that you developed specifically for
beta_Cstill make sense when thinking about comparing raw S values -> my thought is yes because its the largest coverage you can get across all samples.