具有由人和顺序组成的数据框...
person order elements
Alice [drink, snack, salad, fish, dessert] 5
Tom [drink, snack] 2
John [drink, snack, soup, chicken] 4
Mila [drink, snack, soup] 3
我想知道顾客主要吃什么。因此,我想添加另一列[main_meal],使其成为我的df。
person order elements main_meal
Alice [drink, snack, salad, fish, dessert] 5 fish
Tom [drink, snack] 2 none
John [drink, snack, soup, chicken] 4 chicken
Mila [drink, snack, soup] 3 none
规则是,如果客户订购了4餐或更多餐,则意味着第4个元素始终是主菜,因此我想从订单列上的列表中提取第4个元素。如果它包含少于4个元素,则将'main_meal'分配为没有。我的代码:
df['main_meal'] = ''
if df['elements'] >= 4:
df['main_meal'] = df.order[3]
else:
df['main_meal'] = 'none'
它不起作用:
ValueError Traceback (most recent call last)
<ipython-input-100-39b7809cc669> in <module>()
1 df['main_meal'] = ''
2 df.head(5)
----> 3 if df['elements'] >= 4:
4 df['main_meal'] = df.order[3]
5 else:
~\Anaconda\lib\site-packages\pandas\core\generic.py in __nonzero__(self)
1571 raise ValueError("The truth value of a {0} is ambiguous. "
1572 "Use a.empty, a.bool(), a.item(), a.any() or
a.all()."
-> 1573 .format(self.__class__.__name__))
1574
1575 __bool__ = __nonzero__
ValueError: The truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all().
我的代码有什么问题?
答案 0 :(得分:4)
使用str
方法进行切片
In [324]: df['order'].str[3]
Out[324]:
0 fish
1 NaN
2 chicken
3 NaN
Name: order, dtype: object
In [328]: df['main_meal'] = df['order'].str[3].fillna('none')
In [329]: df
Out[329]:
person order elements main_meal
0 Alice [drink, snack, salad, fish, dessert] 5 fish
1 Tom [drink, snack] 2 none
2 John [drink, snack, soup, chicken] 4 chicken
3 Mila [drink, snack, soup] 3 none
答案 1 :(得分:1)
对于小型数据框,可以按照@Zero's solution使用str
访问器。对于较大的数据框,您可能希望使用NumPy表示形式创建一个序列:
# Benchmarking on Python 3.6, Pandas 0.19.2
df = pd.concat([df]*100000)
%timeit pd.DataFrame(df['order'].values.tolist())[3] # 125 ms per loop
%timeit df['order'].str[3] # 185 ms per loop
# check results align
x = pd.DataFrame(df['order'].values.tolist())[3].fillna('None').values
y = df['order'].str[3].fillna('None').values
assert (x == y).all()
答案 2 :(得分:0)
您还可以使用apply
和lambda函数。
df['main_meal'] = df['order'].apply(lambda r: r[3] if len(r) >= 4 else 'none')
对于大型数据集,它的回答要慢于@jpp's,但对于较小的数据集,它的回答要快(且更冗长),比@jpp和@Zero's都要快(请注意,我添加了.fillna()
,因此它们返回了结果相同):
%timeit df['order'].apply(lambda r: r[3] if len(r) >= 4 else 'none') # 242 µs
%timeit pd.DataFrame(df['order'].values.tolist())[3].fillna('none') # 1.17 ms
%timeit df['order'].str[3].fillna('none') # 487 µs
# Large dataset
df = pd.concat([df]*100000)
%timeit df['order'].apply(lambda r: r[3] if len(r) >= 4 else 'none') # 118ms
%timeit pd.DataFrame(df['order'].values.tolist())[3].fillna('none') # 51.8ms
%timeit df['order'].str[3].fillna('none') # 309ms
如果您检查它们的值,它们将匹配。
x = df['order'].apply(lambda r: r[3] if len(r) >= 4 else 'none')
y = pd.DataFrame(df['order'].values.tolist())[3].fillna('none')
z = df['order'].str[3].fillna('none')
(x.values == y.values).all() # True
(x.values == z.values).all() # True
Python 3.6.6 |熊猫0.23.4