將空（ish）字符串轉換爲空的最有效方法

Question

我有一個只包含空字符串和空字符串的pandas系列。我想將它們轉換爲'null'值（例如None）。

def empty_str_to_null(s): 
    """Convert empty strings to None (null)""" 
    s.loc[s.str.strip().str.len() == 0] = None 
    return s 

foo = pd.Series(np.repeat([1,2,3,'',None,np.NaN, ' ', ' a'],1E6)) 

>>> %time bar = empty_str_to_null(foo) 

這個工程，但並不那麼快。

CPU times: user 7.67 s, sys: 260 ms, total: 7.93 s 
Wall time: 8.38 s 

我需要爲許多不同的領域重複這樣做。

有沒有更好的（更快）的方法？

Answer 1

這裏有一個方法 -

def empty_str_to_null_slicer(s): 
    a = s.values.astype(str) 
    # slicer_vectorized from https://stackoverflow.com/a/39045337/ 
    mask = (slicer_vectorized(a,0,1)==' ') | (a=='') 
    s[mask] = None 
    return s 

採樣運行 -

In [245]: s = pd.Series(np.repeat([1,'',' ',None,np.NaN],2)) 

In [246]: s 
Out[246]: 
0  1 
1  1 
2   
3   
4   
5   
6 None 
7 None 
8  NaN 
9  NaN 
dtype: object 

In [247]: a = s.values.astype(str) 
    ...: mask = (slicer_vectorized(a,0,1)==' ') | (a=='') 
    ...: s[mask] = None 
    ...: 

In [248]: s 
Out[248]: 
0  1 
1  1 
2 None 
3 None 
4 None 
5 None 
6 None 
7 None 
8  NaN 
9  NaN 
dtype: object 

運行測試 -

途徑 -

# Original approach 
def empty_str_to_null(s0): 
    s = s0.copy() 
    """Convert empty strings to None (null)""" 
    s.loc[s.str.strip().str.len() == 0] = None 
    return s 

# Proposed approach 
def empty_str_to_null_slicer(s0): 
    s = s0.copy() 
    a = s.values.astype(str) 
    # slicer_vectorized from https://stackoverflow.com/a/39045337/3293881 
    mask = (slicer_vectorized(a,0,1)==' ') | (a=='') 
    s[mask] = None 
    return s 

計時 -

In [228]: foo = pd.Series(np.repeat([1,'',' ',None,np.NaN],1E6)) 

In [229]: %timeit empty_str_to_null(foo) 
1 loop, best of 3: 4.17 s per loop 

In [230]: %timeit empty_str_to_null_slicer(foo) 
1 loop, best of 3: 573 ms per loop