0

I am doing some predictive modeling. Among various methods, I am also working with MARS. I always get a messages "earth glm PA: did not converge after 25 iterations" I tried some tips (such as adding maxit = 200) i got from stackoverflow, but no luck. The model was replicated for 10 times. I get the message for 6 to 10 times. which means, the problem was at worst in all replications.

May I know what will be consequence? and any idea how to resolve it.

 library(sdm)
 WE.mar<- sdm(PA~., data=dt.W,  methods = c("mar"), replication='sub', test.percent=30, n = 10, maxit = 200)

Your suggestions are highly appreciated. Many thank in advance.

MKS
  • 13
  • 1
  • 4
  • Could you be more specific? Please have a look at https://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example to learn how to make a reproducible post. – Marco Sandri May 26 '17 at 11:59
  • Sorry, I could not make it reproducible. It needs more than 20,000 records to occur the issue and I have about 40,000 random points and 184 presence points. At lower number of background points, there is less likely to occur the issue (my experience). Thanks – MKS May 26 '17 at 12:15
  • Hey, just came across this while dealing with the same issue. Yes, you should be concerned about convergence, it means you shouldn't trust the model's parameter estimates. Most models have options for increasing the number of maximum iterations. I'm currently searching for this option for the earth package in R. – Kreitz Gigs Jan 18 '19 at 07:12

0 Answers0