如何在np.ogrid生成的n×n矩阵中替换指定列且保留对角线值？

Got it, let's break this down step by step to solve your problem. The key is to first generate your base grid, then target the specific column you want to modify—while explicitly skipping the diagonal element where a = b to keep its original value.

Step-by-Step Implementation #

First, let's start with a concrete example that aligns with your requirements:

import numpy as np

# Pre-defined geo_mean (replace this with your actual pseudo-random generator logic)
def compute_geo_mean():
    # Generate 50 pseudo-random values and calculate their geometric mean
    rng = np.random.default_rng(seed=42)  # Seed for reproducibility
    random_vals = rng.uniform(1.0, 10.0, 50)
    return np.prod(random_vals) ** (1/50)

geo_mean = compute_geo_mean()
n = 20  # Size of your n×n grid
target_col = 17  # The column you want to replace (b=17)

# Generate the base ogrid arrays (a is column vector, b is row vector)
a, b = np.ogrid[:n, :n]

# Create your base grid with your original exponential formula
# Replace this with your actual base formula (e.g., np.exp(a - b) or whatever you're using)
base_grid = np.exp(a + b)

# Identify rows in the target column where a != b (i.e., non-diagonal positions)
non_diag_rows = np.where(a[:, 0] != target_col)[0]

# Apply your new formula with geo_mean to these non-diagonal positions
# Customize this formula to match your exact needs—this is just an example
base_grid[non_diag_rows, target_col] = np.exp(a[non_diag_rows, target_col] * geo_mean + target_col)

# Verify the diagonal value remains unchanged
print(f"Diagonal value at ({target_col}, {target_col}) is preserved: {base_grid[target_col, target_col]}")

Key Details Explained #

• Ogrid Setup: a and b are broadcast into n×n matrices, so a[i,j] = i and b[i,j] = j for all positions.
• Target Column Isolation: We use np.where to get all row indices where the row number doesn't match the target column index—this skips the diagonal element (target_col, target_col).
• Custom Formula Application: We only overwrite the non-diagonal elements in the target column with your new formula that includes geo_mean. The diagonal element stays as computed by your original base formula.
• Flexibility: Adjust the new formula (the line assigning to base_grid[non_diag_rows, target_col]) to match your exact exponential logic—you can use a, target_col (since b is 17 for this column), and geo_mean however you need.

Edge Cases to Consider #

• If n is smaller than your target column index (e.g., n=15 and target_col=17), add a quick check to avoid index errors:

  if target_col >= n:
      raise ValueError("Target column index is out of bounds for the n×n grid")
  

• If you need to apply this to multiple columns, wrap the logic in a loop over your target column indices.

内容的提问来源于stack exchange，提问作者Luke