如何对两列进行分组，而不考虑值的顺序?

nm = ["val1", "val2"]
grp = df[nm].apply(lambda x: tuple(sorted(list(x))), axis=1)
s = df.val3.groupby(grp).mean()
s.index = pd.MultiIndex.from_tuples(s.index, name=nm)
s.reset_index()
#   val1 val2  val3
# 0    a    b   5.0
# 1    f    k   3.5

另一种解决方案，使用frozenset:

x = (
df.groupby(df[["val1", "val2"]].apply(frozenset, axis=1))
.agg({"val1": "first", "val2": "first", "val3": "mean"})
.reset_index(drop=True)
)
print(x.to_markdown())

打印: