我的解决方案是这样的：增加一个结构来快速找到相反的对，并创建一个布尔掩码进行过滤，而不是在循环中调用drop()。

import pandas as pd

data = pd.DataFrame(
    [
        ["Brown", "Bottle", 1234555, 100],
        ["yellow", "Cup", 1234555, 80],
        ["Red", "Bottle", 1234555, -100],
        ["Red", "Bottle", 1234555, -100],
        ["Brown", "Bottle", 1234533, 100],
    ],
    columns=["A", "B", "C", "D"],
)

# "lookup table"
seen = {} # {(key1, key2): (index, value)}
# which rows to keep?
mask = pd.Series(True, index=data.index)

# itertuples is faster than iterrows
for row in data.itertuples():
    # create a lookup key
    key = (row.B, row.C)
    if key not in seen:
        # store Index and Value in the "lookup table"
        # if we haven't seen this key before
        seen[key] = (row.Index, row.D)
    else:
        prev_index, prev_value = seen[key]
        # if the stored value is the opposite of the current one
        if prev_value == -row.D:
            # we don't want to keep both rows
            mask.loc[prev_index] = False
            mask.loc[row.Index] = False
            # and remove the key from the lookup table
            del seen[key]
        # else:
            # undefined case:
            # the key exists, but the value is not
            # the opposite of the previous one

# remove "collapsed" rows from the data
result = data[mask]