Styjun

I've got a ton of data transforms defined in a batch transform language that supports this structure: x = iif(condition, a, b). I want to rewrite these using dataframes.

I'm using Dataframe.assign() but get ValueError: The truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all().

 import pandas as pd
 df = pd.DataFrame(['apple', 'orange', 'granite'], columns=['name'])
 df.assign(taste = lambda x: 'rocky' if (x.name=='granite') else 'yummy')

---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-39-e9ad71ccc45b> in <module>()
----> 1 df.assign(taste = lambda x: 'rocky' if (x.name=='granite') else 'yummy')

~Anaconda3libsite-packagespandascoreframe.py in assign(self, **kwargs)
 3305 if PY36:
 3306 for k, v in kwargs.items():
-> 3307 data[k] = com._apply_if_callable(v, data)
 3308 else:
 3309 # <= 3.5: do all calculations first...

~Anaconda3libsite-packagespandascorecommon.py in _apply_if_callable(maybe_callable, obj, **kwargs)
 403 
 404 if callable(maybe_callable):
--> 405 return maybe_callable(obj, **kwargs)
 406 
 407 return maybe_callable

<ipython-input-39-e9ad71ccc45b> in <lambda>(x)
----> 1 df.assign(taste = lambda x: 'rocky' if (x.name=='granite') else 'yummy')

~Anaconda3libsite-packagespandascoregeneric.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().

At first I thought this was due to a limitation of the keywords allowed in assign(), but a very similar construct works with apply():

df['name'].apply(lambda x: 'rocky' if (x=='granite') else 'yummy')

0 yummy
1 yummy
2 rocky
Name: name, dtype: object

However, this doesn't allow me use to an if-condition that uses multiple columns from the dataframe. Is there a way to get assign() to work?

When calling Series.apply, the lambda receives each row value (i.e., a scalar value). With assign, the lambda receives the entire DataFrame. Understanding this means you can now do something such as

df.assign(taste=lambda x: np.where(x['name'] == 'granite', 'r', 'y'))

 name taste
0 apple y
1 orange y
2 granite r

Or,

df.assign(taste=np.where(df['name'] == 'granite', 'r', 'y'))
 name taste
0 apple y
1 orange y
2 granite r

Or, more simply, for in-place assignment,

df['taste'] = np.where(df['name'] == 'granite', 'r', 'y')
df

 name taste
0 apple y
1 orange y
2 granite r

assign is not the function you should using with condition assignment

df['taste']=np.where(df['name'].eq('granite'),'rocky','yummy')
df
Out[513]: 
 name taste
0 apple yummy
1 orange yummy
2 granite rocky