gratia overplotting

bad_code.R

# put all yr models in a list
model_list <- list(b_norm, b_term, b_term_sel)
# name them for nice labels
names(model_list) <- c("No selection", "Shrinkage smoother", "Extra penalty")
# all the terms (there's probably a gratia built-in for this)
term_list <- c("s(Depth)", "s(Bottom)", "s(Surface)")

library(gratia)

# pre-storage
plot_dat <- c()

# loop over terms and models
for(this_term in term_list){
  for(i in seq_along(model_list)){
  
    # evaluate this term for this model
    this_smoo <- evaluate_smooth(model_list[[i]], this_term)
    
    # add identifier columns
    this_smoo$model <- names(model_list)[i]
    this_smoo$term <- this_term

    # rename the column with the covariate values
    this_smoo[["covar"]] <- this_smoo[[sub("s\\((.+)\\)", "\\1", this_term)]]
    # remove old column
    this_smoo[[sub("s\\((.+)\\)", "\\1", this_term)]] <- NULL

    # squidge it together
    plot_dat <- rbind.data.frame(plot_dat, this_smoo)
  }
}

# adhoc plot
p_shrinky <- ggplot(plot_dat, aes(x=covar, group=model, fill=model)) +
  geom_ribbon(aes(ymin=est-2*se, ymax=est+2*se), alpha=0.4) +
  geom_line(aes(y=est)) +
  facet_wrap(~term, scale="free") +
  theme_minimal()

produces something like this: