Pandas技术需求：若列最后非零值小于0则替换为0

Great question! Let's build on the forward (top-to-bottom) logic you already used to solve this reverse (bottom-to-top) problem efficiently—no loops required, just clean vectorized pandas operations.

Step 1: Define What We Need to Target #

We only want to replace the very last non-zero value in each column if that value is negative. All other values (including earlier negatives) should stay as-is.

Step 2: Build the Target Mask #

We'll create a boolean mask that marks exactly those positions we need to change:

1. First, flag all non-zero values in the DataFrame:

   non_zero = df != 0
   

2. Find the last non-zero position in each column. We can do this by comparing the cumulative count of non-zeros (row-by-row) to the total number of non-zeros in the column:

   last_non_zero_pos = non_zero.cumsum(axis=0) == non_zero.sum(axis=0)
   

3. Narrow this mask to only include positions where the value is negative:

   target_mask = last_non_zero_pos & (df < 0)
   

Step 3: Apply the Mask to Update Values #

Use pandas' mask() method to set the marked positions to 0:

import pandas as pd

# Your original DataFrame
df = pd.DataFrame(
    {'A': [1,2,-2,0,0], 
     'B': [0, 0, 0, 3, -2], 
     'C' : [0, 0, -2, 4, 0], 
     'D': [0, -3, 2, 1, -2]} 
)

# Create the mask for target positions
non_zero = df != 0
last_non_zero_pos = non_zero.cumsum(axis=0) == non_zero.sum(axis=0)
target_mask = last_non_zero_pos & (df < 0)

# Replace the negative last non-zero values with 0
df_end = df.mask(target_mask, 0)

# Print the result
print(df_end)

Expected Output #

A  B  C  D
0  1  0  0  0
1  2  0  0 -3
2  0  0 -2  2
3  0  3  4  1
4  0  0  0  0

How This Works #

• non_zero.cumsum(axis=0) keeps a running total of non-zeros for each column as we go down the rows.
• Comparing this to non_zero.sum(axis=0) (the total non-zeros per column) gives us a boolean matrix where only the last non-zero row in each column is True.
• We intersect this with df < 0 to filter out any last non-zero values that are positive.
• df.mask() replaces any value where the mask is True with 0, leaving all other data unchanged.

This method is efficient, scalable, and works even if your data doesn't have alternating positive/negative values (unlike the workaround you used for the forward case).

内容的提问来源于stack exchange，提问作者Leo