可以使用numpy.select：

tmp = df.set_index('Seller').eq(0)

df['Seller'].set_axis(
    np.select([tmp.all(axis=1), tmp.iloc[:, -3:].all(axis=1), tmp.any(axis=1)],
              ['Always 0', '0 in last 3 dates', 'Sometimes 0'], 'Never 0')
).groupby(level=0).count()

或者：

pd.Series(np.select([tmp.all(axis=1), tmp.iloc[:, -3:].all(axis=1), tmp.any(axis=1)],
          ['Always 0', '0 in last 3 dates', 'Sometimes 0'], 'Never 0')).value_counts()

输出：

0 in last 3 dates    1
Always 0             2
Never 0              2
Sometimes 0          4
Name: Seller, dtype: int64

如果你想要列表：

df['Seller'].set_axis(
    np.select([tmp.all(axis=1), tmp.iloc[:, -3:].all(axis=1), tmp.any(axis=1)],
              ['Always 0', '0 in last 3 dates', 'Sometimes 0'], 'Never 0')
).groupby(level=0).agg(list)

输出：

0 in last 3 dates                               [SellerE]
Always 0                               [SellerD, SellerI]
Never 0                                [SellerG, SellerH]
Sometimes 0          [SellerA, SellerB, SellerC, SellerF]
Name: Seller, dtype: object