biomod backcasted model vs. cross-validated

bm_fit.R

#... cont'd

#### Backcasting part ####
tic("Formating model data")
cat("Formating data for backcasting...", "\n")

model_data <- BIOMOD_FormatingData(resp.var = presabs[, "PA"], expl.var = rasters_region_stack,
                                   eval.resp.var = presabs_eval[, "PA"],
                                   eval.resp.xy = presabs_eval[, c("x", "y")],
                                   resp.xy = presabs[, c("x", "y")], 
                                   resp.name = model_id, na.rm = TRUE)			   
toc(log = TRUE)
mem_used()

tic("Modeling")
cat("Modeling...", "\n")
model_out <- BIOMOD_Modeling(model_data, models = "GLM", models.options = biomod_options, 
                             VarImport = 100, models.eval.meth = c("TSS", "ROC"),
                             SaveObj = TRUE, rescal.all.models = FALSE, do.full.models = FALSE,
                             modeling.id = model_id)
toc(log = TRUE)
mem_used()

# Extract model parameters with load()
# Or figure out better way
eval_values <- model_out %>% 
  get_evaluations() %>%
  as.data.frame() %>%
  select(contains("Testing.data")) %>%
  mutate(Var = row.names(.)) %>% 
  select(Var, everything())

write_csv(eval_values, str_c(eval_folder, "/Modelout_eval_", model_id, ".csv"))

var_imp_values <- model_out %>% 
  get_variables_importance() %>% 
  as.data.frame() %>%
  mutate(Var = row.names(.)) %>% 
  select(Var, everything())

write_csv(var_imp_values, str_c(eval_folder, "/Modelout_varimp_", model_id, ".csv"))

#### Cross validation ####
tic("Formating model data")
cat("Formating data for crossvalidation...", "\n")
model_data_cv <- BIOMOD_FormatingData(resp.var = presabs[, "PA"], expl.var = rasters_region_stack,
                                   resp.xy = presabs[, c("x", "y")], 
                                   resp.name = paste0(model_id, "_cv"), na.rm = TRUE)
toc(log = TRUE)
mem_used()

tic("Modeling")

cat("Modeling...", "\n")
model_out_cv <- BIOMOD_Modeling(model_data_cv, models = "GLM", models.options = biomod_options, 
                             VarImport = 100, NbRunEval = 1, DataSplit = 67, models.eval.meth = c("TSS", "ROC"),
                             SaveObj = TRUE, rescal.all.models = FALSE, do.full.models = FALSE,
                             modeling.id = paste0(model_id, "_cv"))
toc(log = TRUE)
mem_used()

eval_values_cv <- model_out_cv %>% 
  get_evaluations() %>%
  as.data.frame() %>%
  select(contains("Testing.data")) %>%
  mutate(Var = row.names(.)) %>% 
  select(Var, everything())

write_csv(eval_values_cv, str_c(eval_folder, "/Modelout_eval_", model_id, "_cv.csv"))

var_imp_values_cv <- model_out_cv %>% 
  get_variables_importance() %>% 
  as.data.frame() %>%
  mutate(Var = row.names(.)) %>% 
  select(Var, everything())

write_csv(var_imp_values_cv, str_c(eval_folder, "/Modelout_varimp_", model_id, "_cv.csv"))