wallaceEcoMod / wallace

an interactive, reproducible, expandible, instructional, and open-source GUI-based app for ecological niche modeling
https://wallaceecomod.github.io/
GNU General Public License v3.0
133 stars 46 forks source link

Warning: Error in intI: index larger than maximal 185 #352

Closed Oliveira-D-R closed 2 years ago

Oliveira-D-R commented 3 years ago

After upgrading to the new version of wallace I am encountering a series of errors. The last one that appeared was:

Running ENMeval v2.0.0 with maxnet from maxnet package v0.1.2

|========= | 10%Registered S3 methods overwritten by 'adehabitatMA': method from print.SpatialPixelsDataFrame sp
print.SpatialPixels sp
|======================================================= | 60%Warning: from glmnet Fortran code (error code -186); Convergence for 186th lambda value not reached after maxit=100000 iterations; solutions for larger lambdas returned Warning: Error in intI: index larger than maximal 185

jamiemkass commented 3 years ago

Were you not getting this error before with the previous Wallace version? This is a maxnet error that usually pops up when using hinge features on a big and complex dataset. Not sure why ENMeval 2 would throw this error if ENMeval 0.3.1 did not.

Can you also try models without hinges and report if this goes away? Just a note -- maxent.jar often does not error under these circumstances.

Oliveira-D-R commented 3 years ago

That worked! Thank you very much! It was the hinge. I also didn't understand why I didn't receive this error before the update. I will try to do it again using maxent.jar. Again, thank you very much.

jamiemkass commented 3 years ago

Great! Please let us know if you think ENMeval 2 is causing the error, and we can investigate. I am pretty sure this wouldn't be the case for this particular error though.

gepinillab commented 3 years ago

Hi @jamiemkass,

I decided to reopen this issue again because it was reported on the Google group. https://groups.google.com/g/wallaceEcoMod/c/4ofAiIL3Dbs

It looks like Ricardo Torres got a similar message after updating Wallace and ENMeval to their newest versions. Also, when he did not use Hinge, he got the issue with rangeModelMetadata (not sure if ecospat too).

I do not have a clue about the reason for the first issue. Any thoughts?

Best, Gonzalo

jamiemkass commented 3 years ago

Has anyone confirmed that ENMeval 0.3.1 did not throw this error for maxnet on the same dataset? The only difference in implementation for maxnet is that 2.0.0 sets addsamplestobackground to TRUE for maxnet() to avoid this error (or a similar one).

What was the issue with rangeModelMetadata? This was not in Ricardo's post.

gepinillab commented 3 years ago

Users are reporting this error after updating to Wallace v1.1 (ENMeval 2.0). Ricardo told me via e-mail (no google group) that he did not have this problem using Wallace v1.0.6 (ENMeval 0.3). I am not sure about he was using the same dataset, but it is possible because he was using both versions (Wallace 1.06 and 1.1) on the same date. We can ask him to test an ENMeval branch that has addsamplestobackground = FALSE to see if the problem persists.

About rangeModelMetadata, he got the following message after using maxnet without Hinge: Warning: Error in loadNamespace: there is no package called ‘rangeModelMetadata’ [No stack trace available]

So, I recommended installing RMM and ecospat, which fixed the error when he was not using Hinge. This issue is very similar to the one reported in another issue (https://github.com/wallaceEcoMod/wallace/issues/350).

gepinillab commented 3 years ago

Okay. I already asked Ricardo to test a forked branch with addsamplestobackground= FALSE. Let´s wait to see if he has time to test it.

gepinillab commented 3 years ago

The version of maxnet that adds that parameter is not on CRAN. Ricardo will test if installing the maxnet version on Github fix the problem.

gepinillab commented 3 years ago

Installing maxnet from GitHub is not fixing the problem either. I will ask Bethany to see if we can obtain a reproducible dataset to obtain the same error. Still, I think that someone needs to approach Steven Phillips related to updating maxnet on CRAN.

jamiemkass commented 3 years ago

Okay, thanks. Must be some namespace issue with rangeModelMetadata. It should be installed with the installation of ENMeval.

Concerning maxnet, we do need a reproducible example showing that 0.3.1 does not throw the error yet 2.0.0 does. Would appreciate if someone can investigate. Maybe Ricardo can send you some sample data?

jamiemkass commented 3 years ago

Steven Phillips said he'd make the CRAN update, and I just started dev on ENMeval 2.0.1 to fix some of these issues. Please see the new version on the master branch and try it out.

gepinillab commented 3 years ago

Awesome! Everything looks fine with ENMeval 2.0.1. I will ask Ricardo about the dataset.

gepinillab commented 3 years ago

Ricardo is using his own data, so it is not an option to get a reproducible error. However, Bethany will try to get one next week.