Hi,
I wonder if someone can help me. I have built a gam model to predict the
presence of cold water corals and am now trying to evaluate my model by
splitting my dataset into training/test datasets.
In an ideal world I would use the sample() function to randomly select rows of
data for me so for example with 936 rows of data in my HH dataset I might say
ss <- sample(nrow(HH), size = nrow(HH)-312, replace = FALSE)
training<-HH[ss,]
test<-HH[-ss,]
in order to create a random training sub-sample of roughly 65% of my data and
test of 35%. (I would use a for() loop to automate the process of building the
datasets and running the prediction e.g.1000times)
The problem is that I do have 2 caveats for the subsampling:
a) I need to have control over the prevalence (proportion of observed
presences within the dataset) in my build and test datasets
I realise I could do this by sorting my column of presences and absences and
then taking a subsample of the required size from the rows containing presences
then the rows containing absences and combining them.
e.g. presence_records<-sample(1:117,size=75,replace=FALSE)
absence_records<-sample(118:936,size=549,replace=FALSE)
ss<-c(presence_records,absence_records)
but...
b) My samples are within video transects and due to the risk of
autocorrelation within each transect, ideally it is by transect cluster that
they will be randomly selected. (a point within a transect cannot be allocated
to the training dataset when another point from that same transect is already
allocated to the test dataset)
Is there a way I can fulfil both of these caveats and come out with my
(slightly less)random subsamples?
Many thanks for your time!
All the best,
Bex
[[alternative HTML version deleted]]
______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
