problems with maxent.jar

[hzumbado]

I am having issues running models with ENMeval 2.0.5 using maxent.jar even on different computers. The following model runs perfectly with ENMeval 2.0.4.

mx <-
ENMevaluate(
occs = occs,
envs = envs,
bg = bg,
tune.args =
list(fc =
c('L', 'Q', 'H', 'LQ', 'LH', 'QH', 'LQH'),
rm = 1:4),
partitions = "block",
algorithm = "maxent.jar",
doClamp = TRUE,
overlap = FALSE,
taxon.name = my_species,
parallel = TRUE)

I am able to run this with ENMeval 2.0.5 algorithm = 'maxnet' but it fails with maxent.jar. Specifically the error shows:

*** Running initial checks for Isthmohyla_pseudopuma ... ***

• Clamping predictor variable rasters...
• Model evaluations with spatial block (4-fold) cross validation and lat_lon orientation...

*** Running ENMeval v2.0.5 for Isthmohyla_pseudopuma with maxent.jar v3.4.3 using the predicts package v0.1.19 ***

Error in checkForRemoteErrors(val) : 12 nodes produced errors; first error: method predict with signature (Ljava/lang/String;[Ljava/lang/String;[[I[Ljava/lang/String;)[D not found

Thanks

[wouterbeukema]

I'm running into the same issue. My R version, packages, and Java are all up to date.

[jamiemkass]

Sorry you are having issues. To be compatible with terra, this version uses the predicts package to run maxent.jar instead of dismo, but I was under the impression the implementation was the same (the developer is still Hijmans).

Can you confirm this errors only when run in parallel? I ran checks for this before uploading the pkg, but I will check once more on my machine too.

[hzumbado]

I'm having the issue with both options (parallel = T) and (parallel = F) with ENMeval 2.0.5-2. I am able to run the model using maxent.jar with ENMeval 2.0.4 and dismo, but for some reason I can't use the predict function (either terra or predicts package) to produce the raster of continuous predictions. It shows a weird message about using 'predict's instead of terra even when I am using predicts::predict

[jamiemkass]

Can you please send me the exact error text? It might help me diagnose.

[hzumbado]

Sure! envs is the spatraster. bm = best model

predicts::predict(
bm,
envs,
args = c("outputformat=logistic"))

terra::predict(
bm,
envs,
args = c("outputformat=logistic"))

Both methods produce the same error

Error in .predictSpatRaster(object, x, ext, args, filename, ...) :
Use the 'predicts' package to make MaxEnt models with a SpatRaster

[jamiemkass]

Thank you. And my last request -- can you confirm you've updated terra and predicts to their most recent versions? You may need to update predicts from Github if it's not updated on CRAN yet.

[hzumbado]

I'm using terra 1.8-50 and predicts 0.1-19 (both are the most recent versions)

[jamiemkass]

Okay, so I tried it on my machine, and I cannot reproduce the error. I get a problem when trying to make predictions with MaxEnt model objects if I use categorical variables, because they are internally converted to numeric, but I just realized that MaxEnt() now does not error when given non-numeric categorical variables (maybe due to a recent update?), so I will remove all internal conversion of categoricals to numeric in the next version to avoid this problem.

However, my predict() functions work fine, and no issues with the parallelization. Could this be due to a system issue? Are you on a Mac or PC or other?

[hzumbado]

I'm using only continuous bioclimatic variables from WorldClim and the model runs perfectly with maxnet, the problem perhaps is java related, since maxent.jar produces the error. Honestly, I have no idea. I asked a friend of mine and he's having the same issue. We both are PC users. I am obtaining the same errors with two laptops, HP and DELL, both with Windows 11 Pro. I also reinstalled java and rJava but still got the same error.

[jamiemkass]

Sorry about that, but I think this is some weird issue with the predicts package. I'll do some investigation on a Windows machine and hopefully get back to you. In the meantime, please use maxnet or v2.0.4.