Generate Lag Features

lag.R

lf1 <- function (dataset2, valcol2, numfeats){
  # Computes the lag features for specific column specified by valcol2 within dataset2
  # This creates total numfeats new features and it assumes dataset is already sorted by time
  # Args:
  #   dataset2: dataset to be used for adding new features.
  #   valcol2:  index of the column for which we need to create new features.
  #   numfeats: number of features to create. For example, if this is 3, then it will create 3 new lag features
  #
  # Returns:
  #   New dataset with additional features
  
  previous <- numfeats #total number of features to be generated
  orig_names <- names(dataset2)
  n_rows <- dim(dataset2)[1] #number of rows in dataset
  n_cols <- dim(dataset2)[2] #number of columns in dataset
  
  base<-valcol2 #set column index to be featurized
  
  
  
  for (i in 1:previous) {
  
     
    dataset2[(i+1):n_rows,n_cols+i] <- dataset2[1:(n_rows-i),base] #copy 1:n-1 rows for specific column to 2:n as part of lag
    dataset2[1:i,n_cols+i] <- dataset2[1:i,
                                       if (i==1)  base else base+(n_cols-base)+i-1
                                       ] # now adjust values for the first numfeat rows
  }
  
  
  a <- -1:-previous
  
  new_names <- paste(colnames(dataset2)[valcol2],"lag ",a) #create new column names
  
  
  names(dataset2) <- c(orig_names,new_names) #update dataset with new column names
  
  return (dataset2)
}
                                       
GenerateLagFeatures <- function (dataset1, valcol, numfeats){
  # Computes the lag features for specific column specified by valcol2 within dataset2
  # This creates total numfeats new features and it assumes dataset is already sorted by time
  # Args:
  #   dataset1: dataset to be used for adding new features.
  #   valcol2:  array of string that has names' of all  columns in string format for which we need to create new features.
  #   numfeats: number of features to create. For example, if this is 3, then it will create 3 new lag features
  #
  # Returns:
  #   New dataset with additional features
  
  orig_names <- names(dataset1)
  bases <- which((orig_names %in% valcol)) #find columns that needs to be featurized
  
  if (length(bases) == 0)
  {
    print (paste("No columns matching with ", valcol, " found"))
    # error - no column found with name that is matching in the list of names provided to the function
    return (dataset1)
  }
  #generate features for each column
  for (k in 1:length(bases)) {
    
    base <- bases[k]
    
    
    dataset1<- lf1(dataset1, base, numfeats)
  }
  
  
  return (dataset1)
}