向Numpy二维矩阵指定坐标插入值：行内特定位置添加元素方法

Great question! Inserting elements at row-specific positions in a NumPy array can be a bit tricky because vectorized operations like np.insert work uniformly across rows, but we can handle this by processing each row individually and then combining the results. Here's a step-by-step solution:

Step 1: Define Your Original Matrix and Insertion Specifications #

First, let's start with an example original matrix and the insertion rules you mentioned:

import numpy as np
from collections import defaultdict

# Original 2x3 matrix
original = np.array([[1, 2, 3], [4, 5, 6]])

# Define insertions: (row_index, original_column_index, value_to_insert)
insertions = [
    (0, 0, 100),  # Insert 100 after (0,0)
    (1, 1, 101),  # Insert 101 after (1,1)
    (0, 2, 102)   # Insert 102 after (0,2)
]

Step 2: Group Insertions by Row #

We'll organize our insertions so we can process each row's changes in one go:

# Group insertions by their target row
row_inserts = defaultdict(list)
for row_idx, col_idx, val in insertions:
    row_inserts[row_idx].append((col_idx, val))

Step 3: Process Each Row with Insertions #

For each row, we'll apply its insertions. Important: We process insertions starting from the highest original column index first to avoid shifting positions of elements we haven't inserted after yet.

result_rows = []
for i in range(original.shape[0]):
    current_row = original[i].copy()
    # Get insertions for this row, sorted by column index descending
    row_changes = sorted(row_inserts.get(i, []), key=lambda x: -x[0])
    
    for col_idx, val in row_changes:
        # Insert after the specified column: use position col_idx + 1
        current_row = np.insert(current_row, col_idx + 1, val)
    
    result_rows.append(current_row)

# Combine the modified rows into a list of arrays (or pad to make a 2D array if needed)
result = np.array(result_rows, dtype=object)  # Use object dtype for variable-length rows

Step 4: Check the Result #

Running this code will give you exactly the target matrix you described:

print(result)
# Output:
# [array([  1, 100,   2,   3, 102]) array([  4,   5, 101,   6])]

Notes: #

• If you need a regular 2D NumPy array (all rows same length), you'll need to pad shorter rows with a placeholder (like np.nan) to match the longest row's length. For example:

  max_length = max(len(row) for row in result_rows)
  padded_result = np.array([np.pad(row, (0, max_length - len(row)), constant_values=np.nan) for row in result_rows])
  

• This approach scales well even for larger matrices—you just need to update the insertions list with your specific row, column, and value pairs.

内容的提问来源于stack exchange，提问作者Schneems