使用 pandas 获取事件列表

Question

给定一个动物列表，例如：

animals = ['cat', 'dog', 'hamster', 'dolphin']

和熊猫数据框，df：

id    animals
1     dog,cat
2     dog
3     cat,dolphin
4     cat,dog
5     hamster,dolphin 

我想获得一个新的数据框来显示每种动物的出现情况，例如：

animal    ids
cat       1,3,4
dog       1,2,4
hamster   5        
dolphin   3,5

我知道我可以运行一个循环并准备它，但我有超过 80,000 个单词的列表以及超过 100 万行的数据框，因此使用循环来完成它需要很长时间。有没有更简单快捷的方法来使用数据框获取结果？

Answer 1

让我们试试get_dummies 然后dot

df.animals.str.get_dummies(',').T.dot(df.id.astype(str)+',').str[:-1]
Out[307]: 
cat        1,3,4
dog        1,2,4
dolphin      3,5
hamster        5
dtype: object

如果考虑列表添加reindex

df.animals.str.get_dummies(',').T.dot(df.id.astype(str)+',').str[:-1].reindex(animals)
Out[308]: 
cat        1,3,4
dog        1,2,4
hamster        5
dolphin      3,5
dtype: object

Answer 2

基于 NumPy 的性能。 -

def list_occ(df):
    id_col='id'
    item_col='animals'
    
    sidx = np.argsort(animals)
    s = [i.split(',') for i in df[item_col]]
    d = np.concatenate(s)
    
    p = sidx[np.searchsorted(animals, d, sorter=sidx)]
    C = np.bincount(p, minlength=len(animals))
    
    l = list(map(len,s))
    r = np.repeat(np.arange(len(l)), l)
    v = df[id_col].values[r[np.lexsort((r,p))]]
    
    out = pd.DataFrame({'ids':np.split(v, C[:-1].cumsum())}, index=animals)
    return out

示例运行 -

In [41]: df
Out[41]: 
  id          animals
0  1          dog,cat
1  2              dog
2  3      cat,dolphin
3  4          cat,dog
4  5  hamster,dolphin

In [42]: animals
Out[42]: ['cat', 'dog', 'hamster', 'dolphin']

In [43]: list_occ(df)
Out[43]: 
               ids
cat      [1, 3, 4]
dog      [1, 2, 4]
hamster        [5]
dolphin     [3, 5]

基准测试

使用给定的样本并简单地增加项目的数量。

# Setup
N = 100 # scale factor
s = [i.split(',') for i in df['animals']]
df_big = pd.DataFrame({'animals':[[j+str(ID) for j in i] for i in s for ID in range(1,N+1)]})
df_big['id'] = range(1, len(df_big)+1)
animals = np.unique(np.concatenate(df_big.animals)).tolist()
df_big['animals'] = [','.join(i) for i in df_big.animals]
df = df_big

时间安排 -

# Using given df & scaling it up by replicating elems with progressive IDs
In [9]: N = 100 # scale factor
   ...: s = [i.split(',') for i in df['animals']]
   ...: df_big = pd.DataFrame({'animals':[[j+str(ID) for j in i] for i in s for ID in range(1,N+1)]})
   ...: df_big['id'] = range(1, len(df_big)+1)
   ...: animals = np.unique(np.concatenate(df_big.animals)).tolist()
   ...: df_big['animals'] = [','.join(i) for i in df_big.animals]
   ...: df = df_big

# @BEN_YO's soln-1
In [10]: %timeit df.animals.str.get_dummies(',').T.dot(df.id.astype(str)+',').str[:-1]
163 ms ± 2.94 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

# @BEN_YO's soln-2
In [11]: %timeit df.animals.str.get_dummies(',').T.dot(df.id.astype(str)+',').str[:-1].reindex(animals)
166 ms ± 4.52 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

# @Andy L.'s soln
%timeit (df.astype(str).assign(animals=df.animals.str.split(',')).explode('animals').groupby('animals').id.agg(','.join).reset_index())
13.4 ms ± 74 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [12]: %timeit list_occ(df)
2.81 ms ± 101 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Answer 3

使用str.split、explode 和agg.join

df_final = (df.astype(str).assign(animals=df.animals.str.split(','))
                          .explode('animals').groupby('animals').id.agg(','.join)
                          .reset_index())

Out[155]:
   animals     id
0      cat  1,3,4
1      dog  1,2,4
2  dolphin    3,5
3  hamster      5