HeleneBlt
commented
2 weeks ago Hello there !
Which version of biomod2 did you use ? You can see it with the command sessionInfo().
On version 4.2-5-2 of biomod2, we have a little problem with the message ! No data has been set aside for modeling evaluation which appears when we check the PAtable independently of the evaluation data. We will correct it on the github version 😄
However, you can check your object myBiomodData_1: it must contain a slot with the evaluation data (or myBiomodData_1@has.data.eval must be TRUE).
The argument models.pa is used to define which pseudo-absence datasets are to be used for each algorithm. Here, as you have only one PA dataset, it is not really useful here. So you can remove it :
myBiomodModelOut_1 <- BIOMOD_Modeling(bm.format = myBiomodData_1,
models = c('MAXNET',"XGBOOST"),
CV.strategy = "user.defined",
CV.user.table = myCVtable,
prevalence = 0.5, var.import = 5,
OPT.strategy = 'default',
metric.eval = c('TSS','ROC'),
scale.models = F,
CV.do.full.models = T,
modeling.id = paste(myRespName,"Now",sep=""))But if you want to use it later, your object must be a list. Example:
AllModelsPA <- list("MAXNET" = "PA1", "XGBOOST" = "PA1")
Hope it helps ! Hélène
Hello, Recently, while working with this package, I encountered a problem. I randomly split the test and training data into 20% and 80%. I also generated 10,000 separate pseudo-absence data points. I defined these settings in the BIOMOD_FormatingData section, but I received the following message:
I want the test data to be independent Thanks in advance for your help. -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= zisp Data Formating -=-=-=-=-=-=-=-=-=
!!! Some data are located in the same raster cell. Only the first data in each cell will be kept as
filter.raster = TRUE.Checking Pseudo-absence selection arguments...
! Some NAs have been automatically removed from your data ! No data has been set aside for modeling evaluation ! Some NAs have been automatically removed from your data -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Done -=-=-=-=
Load current climate layers
now <- list.files(pattern='.tif$', path='mask/var', full.name=T) layers1 <- stack(now) layers1 names(layers1)
---------
Hierarchical Clustering (Multicollinearity)
b <- removeCollinearity( layers1, multicollinearity.cutoff = 0.7, select.variables = T, sample.points = TRUE, nb.points = 10000, plot = TRUE, method = "pearson")
------------------
------------------- "Train Data"---- 80% Data presence and pseudo-absence
TrainData <- 'Train'
myRespName <- 'zisp'
TrainData <- read.csv('Train.csv')
head(TrainData) tail(TrainData) nrow(TrainData)
myResp <- as.numeric(TrainData[,myRespName]) myResp myRespXY <- TrainData[,c("longitude","latitude")] myRespXY
----
------
---------- The custom pseudo-absence table defined.
----- Transform true absences into potential pseudo-absences
myResp.PA <- ifelse(myResp == 1, 1, NA) myResp.PA
myResp.PA.vect <- vect(cbind(myRespXY, myResp.PA), geom = c("longitude","latitude")) myResp.PA.vect
user.defined method
myPAtable <- data.frame(PA1 = ifelse(myResp == 1, TRUE, FALSE)) for (i in 1:ncol(myPAtable)) myPAtable[sample(which(myPAtable[, i] == FALSE), 8000), i] = TRUE
myPAtable nrow(myPAtable) head(myPAtable) tail(myPAtable)
-------------------
-------------------- "Test Data"---- 20% Data presence and pseudo-absence
TestData <- 'Test'
myRespName <- 'zisp'
TestData <- read.csv('Test.csv')
head(TestData) tail(TestData) nrow(TestData)
evalmyResp <- as.numeric(TestData[,myRespName])
evalmyRespXY <- TestData[,c("longitude","latitude")]
---------------------------------------------------------
jar <- paste(system.file(package="dismo"), "maxent.jar", sep='')
----
-----------------RUN_1_Current !!!!
'
'
'
' Format Data with pseudo-absences : user.defined method
' split randomly (80% calibration & 20% evaluation) : user.defined method
------------------------------------------------------------------------------
myBiomodData_1 <- BIOMOD_FormatingData(resp.var = myResp.PA, eval.resp.var = evalmyResp, expl.var = layers1, eval.expl.var = layers1, resp.xy = myRespXY, eval.resp.xy = evalmyRespXY, resp.name = myRespName, filter.raster = T, PA.strategy = 'user.defined', PA.user.table = myPAtable)
myBiomodData_1
-----------------------------------------
------------
AllModels <- c('ANN', 'CTA', 'FDA', 'GAM.mgcv.bam', 'GBM', 'GLM', 'MARS', 'MAXENT', 'MAXNET', 'RF', 'SRE', 'XGBOOST')
myBiomodOption_1 <- bm_ModelingOptions(data.type = 'binary', models = AllModels, strategy = 'default', bm.format = myBiomodData_1, )
myBiomodOption_1
----------Bild Modeling Options
myBiomodModelOut_1 <- BIOMOD_Modeling(bm.format = myBiomodData_1, AllModels, models = c('MAXNET',"XGBOOST"), CV.strategy = "user.defined", CV.user.table = myCVtable, prevalence = 0.5, var.import = 5, OPT.strategy = 'default', metric.eval = c('TSS','ROC'), scale.models = F, CV.do.full.models = T, modeling.id = paste(myRespName,"Now",sep=""))
myBiomodModelOut_1
---------------
Another question I had, how should I define independent test data in the BIOMOD_Modeling section, CV.user.table = myCVtable, when I separated the training and testing data?
Now I run BIOMOD_Modeling, I encounter this error:
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Build Single Models -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Checking Models arguments... Error in .fun_testIfInherits(TRUE, "models.pa", models.pa, "list") : models.pa must be a 'list' object
Best,