求实现矩阵各列单独向上移位并补零的R代码

Hey Olga, let's fix this matrix shifting issue you're facing! I know you've tried several approaches already, so let's break down two practical methods that'll get you the exact output you need.

First, let's make sure we're on the same page with an example. Suppose your original matrix a looks like this:

a <- matrix(1:12, nrow = 4, ncol = 3)
print(a)
#      [,1] [,2] [,3]
# [1,]    1    5    9
# [2,]    2    6   10
# [3,]    3    7   11
# [4,]    4    8   12

Your target matrix b should have the first column unchanged, second column shifted up by 1, third column shifted up by 2, with zeros filling the empty spots:

#      [,1] [,2] [,3]
# [1,]    1    6   11
# [2,]    2    7   12
# [3,]    3    8    0
# [4,]    4    0    0

Method 1: Apply a Custom Shift Function to Each Column #

This approach uses apply() to process each column individually, with a simple helper function to handle the shift and zero padding:

# Define a helper function to shift a column upward by n steps
shift_up <- function(col, n) {
  if (n == 0) return(col)  # No shift for first column
  # Take elements starting from (n+1)th position, then add n zeros at the end
  c(col[(n + 1):length(col)], rep(0, n))
}

# Apply the function to each column: shift count = column index - 1
b <- apply(a, 2, function(col, idx) shift_up(col, idx - 1), idx = 1:ncol(a))

This is clean and easy to read—each column gets shifted exactly the number of steps you need, with zeros filling the bottom of the column where elements were shifted out.

Method 2: Vectorized Indexing (Faster for Large Matrices) #

If you're working with big matrices, this vectorized approach will be more efficient than looping or apply():

n_rows <- nrow(a)
n_cols <- ncol(a)

# Create an index matrix where each position (i,j) points to the original row i + (j-1)
idx <- outer(1:n_rows, 1:n_cols, function(i, j) i + (j - 1))

# Replace indices that exceed the number of rows with NA, then convert NAs to 0
b <- a[idx]
b[is.na(b)] <- 0

# Restore the matrix dimensions
dim(b) <- c(n_rows, n_cols)

Here, we're directly mapping each position in the target matrix to the correct element in the original matrix. Any positions that would go beyond the original matrix's rows get filled with 0 automatically.

Both methods will give you the exact shifted matrix you're looking for. Give them a try—they should work better than the approaches you've tested so far!

内容的提问来源于stack exchange，提问作者Olga