获取数据框列中一系列元素的第一次出现的索引

时间:2017-07-11 17:55:19

标签: python pandas dataframe

考虑一个python数据帧

A      B         C  
1      random    imp1  
2      random    imp2  
5      random    imp3   
1      yes       ---  
2      yes       ---   
3      no        ---   
4      no        ---  
5      yes       ---  

每当列B的值为yes时,我想获取A的值。并且最终对于A的那些值,当这些值在A中出现时,我想要C。所以在这种情况下,我最终想要imp1,imp2和IMP3。

这是否有一种优雅的方式。

3 个答案:

答案 0 :(得分:2)

您可以先使用boolean indexing loc,然后使用duplicated,最后使用isin过滤值a

a = df.loc[df['B'] == 'yes', 'A']
df = df.drop_duplicates('A')
df = df.loc[df['A'].isin(a), 'C']
print (df)
0    imp1
1    imp2
2    imp3
Name: C, dtype: object

<强>计时

np.random.seed(123)
N = 1000000

df = pd.DataFrame({'B': np.random.choice(['yes','no', 'a', 'b', 'c'], N),
                   'A':np.random.randint(1000, size=N),
                   'C':np.random.randint(1000, size=N)})
print (df)

print (df[df.A.isin(df[df.B == 'yes'].A)].drop_duplicates('A').C)
print (df[df.A.isin(df[df.B == 'yes'].drop_duplicates('A').A)].C)

def fjez(df):
    a = df.loc[df['B'] == 'yes', 'A']
    df = df.drop_duplicates('A')
    return  df.loc[df['A'].isin(a), 'C']

def fpir(df):
    a = df.A.values
    b = df.B.values == 'yes'
    d = df.drop_duplicates('A')
    return d.C[np.in1d(d.A.values, a[b])]


print (fjez(df))
print (fpir(df))
In [296]: %timeit (df[df.A.isin(df[df.B == 'yes'].A)].drop_duplicates('A').C)
1 loop, best of 3: 226 ms per loop

In [297]: %timeit (df[df.A.isin(df[df.B == 'yes'].drop_duplicates('A').A)].C)
1 loop, best of 3: 185 ms per loop

In [298]: %timeit (fjez(df))
10 loops, best of 3: 156 ms per loop

In [299]: %timeit (fpir(df))
10 loops, best of 3: 87.1 ms per loop

答案 1 :(得分:2)

让我们使用这个单行:

df[df.A.isin(df[df.B == 'yes'].A)].drop_duplicates('A').C

输出:

0    imp1
1    imp2
2    imp3
Name: C, dtype: object

答案 2 :(得分:1)

这应该非常快

a = df.A.values
b = df.B.values == 'yes'
d = df.drop_duplicates('A')
d.C[np.in1d(d.A.values, a[b])]

0    imp1
1    imp2
2    imp3
Name: C, dtype: object

超越顶级方法。比我的其他方法快50%左右。

from numba import njit

@njit
def proc(f, m):
    mx = f.max() + 1
    a = [False] * mx
    b = [0] * mx
    z = [0] * f.size

    for i in range(f.size):
        x = f[i]
        y = m[i]
        b[x] += 1
        z[i] = b[x]
        a[x] = a[x] or y

    return np.array(z) == 1, np.array(a)[f]

df.C[np.logical_and(*proc(pd.factorize(df.A.values)[0], df.B.values == 'yes'))]

0    imp1
1    imp2
2    imp3
Name: C, dtype: object