如何在Pandas中实现数值符号变化后重置累积和（cumsum）？

Great question! Avoiding iterrows() is always a smart move for better performance, especially with larger DataFrames. Here's a fully vectorized approach that achieves your desired custom cumulative sum without any loops:

Full Code Implementation #

import numpy as np
import pandas as pd

# Generate sample data (matches your example for verification)
df = pd.DataFrame({'data': [-2, -1, 1, -3, -1, 2, 0, 3, -1, -2]})

# Core vectorized logic
# 1. Compute signs, forward-fill zeros with the last non-zero sign
signs = np.sign(df['data'])
signs = signs.replace(0, np.nan).ffill().fillna(0)  # Handle edge case of leading zeros

# 2. Identify where the sign changes from the previous row
change_points = signs != signs.shift(1)

# 3. Assign unique group IDs to consecutive rows with the same sign
groups = change_points.cumsum()

# 4. Calculate cumulative sum within each group
df['custom_cumsum'] = df.groupby(groups)['data'].cumsum()

print(df)

Output (Matches Your Example) #

data  custom_cumsum
0    -2             -2
1    -1             -3
2     1              1
3    -3             -3
4    -1             -4
5     2              2
6     0              2
7     3              5
8    -1             -1
9    -2             -3

How It Works #

Let's break down each step:

• Handle Signs & Zeros: np.sign() gives us the sign of each value (-1, 0, 1). We replace zeros with NaN and use ffill() to carry forward the last non-zero sign—this ensures zeros don't trigger a reset, which aligns with your example where the 0 continues the positive cumulative sum.
• Detect Sign Changes: By comparing each sign to the previous row's sign (shift(1)), we flag every position where the sign switches (these are our reset points).
• Create Group IDs: Using cumsum() on the change points assigns a unique ID to each continuous block of rows with the same sign. Every sign change increments the group ID.
• Grouped Cumulative Sum: groupby(groups) lets us compute the cumulative sum separately for each sign group, automatically resetting the sum whenever the group ID changes.

This method leverages pandas' optimized, C-backed operations, making it significantly faster than iterrows() for large datasets.

内容的提问来源于stack exchange，提问作者Chillar Anand