Predictions in ENMeval

[Madeleine Noll]

Hello, 

I was wondering if anyone knows if it is possible to get a prediction for each fold when using the ENMevaluate function. I have used spatial block cross-validation, but would like a prediction for each of the folds to see the differences between them. 

Or, is it possible to run ENMevaluate with a test dataset which contains both occurrence and background data. At the moment I can only see an option to run with an occurrence dataset. 

Thank you for your help. 

[Jamie M. Kass]

Madeleine,

Based on the way ENMeval is structured, it does not return predictions from folds. This is because the folds are used solely to derive averages of model performance, and these are used to select optimal model settings to make a final model with using the full dataset. In other words, the evaluation stats are used to determine how optimal the settings are for a full model. If your spatial CV stats are good, it means that those settings (fc, rm) result in models with good transferability, so it makes sense to use them to make a final model with the full dataset and use this as your range prediction.

However, an alternate method is to average the predictions of folds and use this as the range prediction. I have seen some studies do this. I think this may be a valid strategy when doing multiple repetitions of random CV, as the folds will not differ too much and the whole process is kind of like bootstrapping. However, for spatial CV, the folds can be very different, and thus predicting some of them will present very difficult transfer exercises, resulting in poor stats for some withheld folds and better stats for others. Averaging these (some very predictions, some better ones) doesn't make much sense to me. Happy to keep discussing this though.

In sum, ENMeval does not return these to you, but you can simply make them on your own by selecting all the data in the training folds (occurrences and background) and running ENMeval with just these.

Jamie

On Wednesday 7 February 2024 at 10:03:37 UTC+9 ebisaus...@gmail.com wrote:

Thank you for your help 

--
You received this message because you are subscribed to the Google Groups "Maxent" group.
To unsubscribe from this group and stop receiving emails from it, send an email to maxent+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/26187c13-5710-407d-ae9e-cd515d012711n%40googlegroups.com.