Update Surr_Train.R for safer Residuals export + export of error indexes.

R/Surr_Train.R

@@ -1307,26 +1307,24 @@ WriteModelResiduals <- function(write_filepath_prefix, training_samples, seed) {
                                         training_samples),
                                       collapse="_"),collapse="")
   }
-  ## write out raw residuals for the best model combination
-  all_res <<- as.data.table(Residuals)
-  ## write names that mean something, so that you know what errors you are looking at 
-  colnames(all_res) <- paste ( dt2v( Sperf, "output" ),
+  # for col length safety, extract first the vectors from Residuals list, then data.table
+  best_res <<- as.data.table(Residuals[ perftop[,"model_id"] ])
+  
+  # write names that mean something, so that you know what errors you are looking at 
+  colnames(best_res) <- paste ( dt2v( Sperf, "output" ),
                                dt2v( Sperf, "method" ),
                                dt2v( Sperf, "preprocessing" ),
-                               sep="_" )
-  
-  best_res <<- subset(all_res,select=perftop[,"model_id"])    
+                               sep="_" )[ perftop[,"model_id"] ] # use only the names needed
   ## write all for testing and manual overview
   #fwrite( all_res, "debug/all_residuals.csv" )
   ## write out only for those that have non-zero columns
   fwrite( SelectActiveColumns(best_res), file = paste0(ensemble_name,"_residuals_best.csv" ))
   ## residuals are errors on validation data set - created as inverse from these indexes
-  fwrite
+  fwrite( as.data.table(inds), file = paste0(ensemble_name,"_train_inds.csv" ) )
   ## err index export
   err_inds = as.data.table( seq(nrow(Fout))[-inds] ) 
   colnames( err_inds ) <- "err_inds"
   fwrite( as.data.table(err_inds), file = paste0(ensemble_name,"_err_inds.csv" ) )
-  
 }
 
 ## once we have screened all the desired possibilities, we can try to