Improved codegen for tm workflow

NEWS.md

@@ -1,8 +1,9 @@
 # sasctl (development version)
 
 * Added `scr_batch` argument to `format_data_json`, allowing to have all data in a single json payload for Viya 2024.7 SCR batch scoring feature.
-* Fixed some issues related to the PMML payload when uploading models to SAS Model Manager
+* Fixed some issues related to the PMML payload when uploading models to SAS Model Manager in some SAS Viya versions
 * Fixed `openBrowser` using a function instead of the argument.
+* Improved `codegen` for Tidy Models workflows guessing the predictors from the pre processing pipeline instead of the model workflow recipe.
 
 # sasctl 0.7.4
 

R/codegen.R

@@ -3,7 +3,7 @@
 #' 
 #' @description
 #' 
-#' \strong{EXPERIMENTAL STATE - MAY NO WORK AS INTENDED}
+#' \strong{EXPERIMENTAL STATE - MAY NOT WORK AS INTENDED}
 #' 
 #' Score code will only be generated successfully for supported models. 
 #' Other models and frameworks will be added in due time. 
@@ -154,15 +154,18 @@ codegen.glm <- function(model, path = "scoreCode.R", rds = "model.rds", cutoff =
 
 codegen.workflow <- function(tm_workflow, path = "scoreCode.R", rds = "model.rds", inputs = NULL,
                              libs = c(), referenceLevel = NULL) {
-
-    predictors <- colnames(tm_workflow[["pre"]][["mold"]][["predictors"]])
-
+
   if (!is.null(inputs)) {
 
     if (!is.vector(names(hmeqTrain))) stop("inputs must be a vector names")
 
     predictors <- inputs
 
+  } else {
+
+    predictorsTable <- tm_workflow[['pre']][['actions']][['recipe']][['recipe']][['var_info']] 
+    predictors <- predictorsTable[predictorsTable$role == "predictor",][["variable"]]
+
   }
 
   target <- colnames(tm_workflow[["pre"]][["mold"]][["outcomes"]])