About "ORmin_" and "ProbMin"

[LuzReyesE]

Hi!
I am implementing the proposed code, but I have a question ..

In "results" I do not see the columns "ORmin_" or "ProbMin". Therefore, when exercising the functions:

resultsData <- acherENM @ results
resultsData% <>% select (settings, contains ('ORmin_'), contains ('ProbMin'))
probs <- resultsData%>% select (contains ('ProbMin'))%>% unlist (., use.names = TRUE)
obsState <- resultsData%>% select (contains ('ORmin_'))%>% unlist (., use.names = TRUE)

"probs" and "obsState" give me null values.

What are you calling in this function? In the case of "ORmin_" it seems to me that it is the minimize ommission rate which is indicated as "avg.test.orMTP" in the ENMevaluate output. But as for "ProbMin", I can't locate the information you mean.

Thanks.

[jmbarrios]

Hi @LuzReyesE, these values are not visible from the ENMEval package as you already notice.

It is correct, ORmin is avg.test.orMTP but the ProbMin values are not easy to get.

ProbMin value is the minimum training presence area but to get this value it is necessary to catch the MaxEnt results and extract it.

A modified ENMEval package can be found on CONABIO/ENMeval. The forked package is outdated but if you look at R/tuning.R file you can find how the ProbMin are extracted from the MaxEnt results.

I might look to an update to that package but I can't give you an ETA.

[LuzReyesE]

Thank you very much for your answer...

I will review the code you share with me...
Regarding what you mention, for a moment I thought that the area predicted as present when the i-th record is removed would be equivalent to the "AUC.bin" part of the ENMevaluate output when "bin.output = T". But it is not logical.
What I do believe is correct is that the part of "test.orMTP_bin" refers to the success-failure variable by excluding the i-th record. However, this data alone is not sufficient to implement the jackknife test as an evaluation metric.
It seems to me that I will select the model configurations with the AICc metric, then with those configurations I will generate models with n-1 of the records to extract the Fractional predicted area values. The modeling with the ENMevaluate and the extraction with the code that you recommend.

Thanks.