passing maxit to glm.control when fitting earth

[elad663]

I am fitting an earth classification model but I need more than 25 iterations. earth calls to glm. glm has a control variable which is passed to glm.control and that is how maxit is passed:

    mars_fit <- earth(formula = response ~ x1 + x2, data = dat,
                      glm = list(family=binomial, control = list(maxit = 50)))

but when I try to pass glm = list(family=binomial, control = list(maxit = 50)) to train, it is not passed through and there is no error message.

Minimal, reproducible example:

it is hard to simulate data that needs more than 25 iterations (default), so I am trying to set max_iter to 1 and show that model summary states a higher value. See information printed under GLM (family binomial, link logit)

Adding glm to the tuneGrid fails with Error: The tuning parameter grid should not have columns nprune, degree.

How can I set maxit?
Thank you.
also: https://stackoverflow.com/questions/55210511/how-to-pass-glm-control-argument-for-earth-using-caret-maxit

Minimal dataset:

rm(list=ls())
set.seed(123)
library(caret)
library(earth)

n=1000
b1 = .72; b2 = -.86; b12 = 2.91
x1 <- rnorm(n = n, sd = 1)
x2 <- rnorm(n = n, sd = 1)
noise <- rnorm(n = n, sd = 3)
prob <- plogis(b1*x1 + b2*x2 + b12*x1^2*x2+ noise)
y <- rbinom(n = n, size = 1, prob)
dat <- data.frame(y=y, x1=x1, x2=x2)

Minimal, runnable code:

fit_control <- trainControl(method = "cv", number = 10)
mars_grid <- expand.grid(degree=1:2, nprune=2:10)

mars_fit <- train(factor(y)~x1+x2, method='earth', trControl = fit_control, tuneGrid=mars_grid, data=dat,
                  glm = list(family=binomial, control = list(maxit = 1)))


> summary(mars_fit$finalModel)
Call: earth(x=matrix[1000,2], y=factor.object, keepxy=TRUE, glm=list(family=function.object),
            degree=2, nprune=10)

GLM coefficients
                                            1
(Intercept)                         0.0133392
h(x1- -0.445662)                    3.8482523
h(-0.10828-x1)                     -2.3430108
h(x1- -0.10828)                    -5.8111977
h(-0.34965-x1) * h(x2- -0.805698)   3.5749689
h(x1- -0.34965) * h(x2- -0.805698) -2.5430102
h(-0.10828-x1) * h(x2-0.366114)    -2.0382430
h(x1- -0.10828) * h(x2-0.686135)    2.5903118
h(x1-0.144476) * h(x2- -0.805698)   5.5678739
h(x1-0.431099) * h(x2-0.76164)     -6.5535434

Earth selected 10 of 17 terms, and 2 of 2 predictors
Termination condition: Reached nk 21
Importance: x1, x2
Number of terms at each degree of interaction: 1 3 6
Earth GCV 0.2073745    RSS 197.7424    GRSq 0.1720774    RSq 0.2089513

GLM (family binomial, link logit):
  nulldev  df      dev  df devratio  AIC iters converged
 1386.194 999 1128.068 990    0.186 1148     5         1

Session Info:

R version 3.5.2 (2018-12-20)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 16.04.6 LTS

Matrix products: default
BLAS: /usr/lib/libblas/libblas.so.3.6.0
LAPACK: /usr/lib/lapack/liblapack.so.3.6.0

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C               LC_TIME=en_US.UTF-8       
 [4] LC_COLLATE=en_US.UTF-8     LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                  LC_ADDRESS=C              
[10] LC_TELEPHONE=C             LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] earth_4.7.0        plotmo_3.5.2       TeachingDemos_2.10 plotrix_3.7-4     
[5] caret_6.0-81       ggplot2_3.1.0      lattice_0.20-38   

loaded via a namespace (and not attached):
 [1] Rcpp_1.0.0         pillar_1.3.1       compiler_3.5.2     gower_0.2.0       
 [5] plyr_1.8.4         iterators_1.0.10   class_7.3-15       tools_3.5.2       
 [9] rpart_4.1-13       ipred_0.9-8        lubridate_1.7.4    tibble_2.0.1      
[13] nlme_3.1-137       gtable_0.2.0       pkgconfig_2.0.2    rlang_0.3.1       
[17] Matrix_1.2-16      foreach_1.4.4      rstudioapi_0.9.0   prodlim_2018.04.18
[21] e1071_1.7-0.1      withr_2.1.2        dplyr_0.8.0.1      stringr_1.4.0     
[25] generics_0.0.2     recipes_0.1.4      stats4_3.5.2       nnet_7.3-12       
[29] grid_3.5.2         tidyselect_0.2.5   glue_1.3.1         data.table_1.12.0 
[33] R6_2.4.0           survival_2.43-3    lava_1.6.5         reshape2_1.4.3    
[37] purrr_0.3.1        magrittr_1.5       splines_3.5.2      MASS_7.3-51.1     
[41] scales_1.0.0       codetools_0.2-16   ModelMetrics_1.2.2 assertthat_0.2.0  
[45] timeDate_3043.102  colorspace_1.4-0   stringi_1.4.3      lazyeval_0.2.1    
[49] munsell_0.5.0      crayon_1.3.4      

[topepo]

It's a bug. I'll fix for the next release.

[topepo]

This should do it but please test. If you don't pass glm then it populates it but passing it overrides the default.

[elad663]

Works well for me. Thank you.