如何根据多个条件在df中创建新列？

Question

I have a df with 3 columns: v1, v2, v3;where

v1=[a,b,c,a] 
v2=[d,d,f,n] 
v3=[a,k,i,j] 

What I like to do is to create new columns based on conditions in column v1~v3.

I can do single condition,

df['v1_a']=np.where(df['v1']=='a',1,0)

it gives a new column named 'v1_a' with 1/0

However, if I want to create a new column based on multiple conditions, this does not work:

df['v2_flag']=np.where(df['v2']=='f' or df['v2']=='h',1,0)

How can I accomplish this?

Answer 1

If you use multiple condition you'll get the following ValueError because np.where() doesn't accept multiple condition :

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

So in your I suggest to use np.logical_or .

df['v2_flag']=np.where(np.logical_or(df['v2']=='f',df['v2']=='h'),1,0)

See the following example too:

>>> a=np.array([2,2,2,5,7,8,1,4,2,3,4,5,6])
>>> np.where(np.logical_or(a==5,a==2),a,0)
array([2, 2, 2, 5, 0, 0, 0, 0, 2, 0, 0, 5, 0])

Answer 2

In python and and or can only give a single result and can't be overridden to have other purposes by modules like the giant row by row comparison you're trying to do.

You need to use the symbolic & (and) and | (or), which are normally used for bit-wise comparisons. These have been re-purposed by pandas to be a row by row comparison, which actually makes sense as being analogous to bit-wise comparisons. That is more of a happy coincidence though, as these were mainly used because these can be overridden by the modules.

Because of the priority of these and equalities, you'll need parentheses around each term or else it would calculate the | before the == which isn't what you want. You can use something like this:

df['v2_flag']=np.where((df['v2']=='f')|(df['v2']=='h'),1,0)

Answer 3

df['v2']=='f' or df['v2']=='h' raises the ValueError before it gets to np.where . The or causes Python to evaluate df['v2']=='f' and df['v2']=='h' in a boolean context. But Pandas Series , like NumPy arrays, refuse to be reduce to a single boolean value -- they raise a ValueError instead .

To fix your code, you could use

df['v2_flag'] = np.where( (df['v2']=='f') | (df['v2']=='h'), 1, 0)

The | performs bitwise-or element-wise over the two boolean-valued Series.

Other ways to define df['v2_flag'] include

df['v2_flag'] = ((df['v2']=='f') | (df['v2']=='h')).astype(int)

or

df['v2_flag'] = df['v2'].isin(['f', 'h']).astype(int)