search

ConsBiol-unibern / SDMtune

Performs Variables selection and model tuning for Species Distribution Models (SDMs). It provides also several utilities to display results.
https://consbiol-unibern.github.io/SDMtune/
Other
23 stars 8 forks source link

[Bug]: JAVA out of memory #40

Open nhill917 opened 1 month ago

nhill917 commented 1 month ago

Describe the bug

When running SDMs with large numbers of presence points, I am getting a JAVA out of memory error from MaxEnt. Is there a way to overcome this? Also, when using the optimizeModel() function, what are reasonable starting values across the different methods (maxent, random forest, boosted regression tree)? I am running a number of SDMs across the various methods and would like to make sure the optimisation step is appropriate. For example, in your vignette you use the below values for MaxEnt, what would you use for random forest and boosted regression tree? h <- list(reg = seq(0.2, 5, 0.2), fc = c("l", "lq", "lh", "lp", "lqp", "lqph"))

Thank you for your help. Kind regards, nick.

Steps to reproduce the bug

library(SDMtune)
NA

Session information

NA

Additional information

NA

Reproducible example

rogerio-bio commented 1 month ago

Hi @nhill917,

Most memory problems in JAVA are because your PC does not have enough RAM to run what the model is requiring. I can easily run a model with 6k presences with 32 variables with 32 GB of RAM, but it takes a while. How big is your dataset?

nhill917 commented 1 month ago

Hi rogerio, yes I think this is the problem. I am running it with 13 variables and for some species beyond 100000 presence points. It is reporting use of 5-6gb of ram. Is there anyway to bypass this or does maxnet bypass java? From memory I tried a run with Maxnet and had similar issues. Thank you for getting back to me.

rogerio-bio commented 4 weeks ago

Hi @nhill917 ,

You can increase the RAM of your PC to meet the requirements of your model, or decrease the number of presence points. Alternatively, consider using a cloud PC for the analysis. This allows you to "rent" the amount of RAM you need to run the model in the cloud.