There is another way without np.where or pd.Series.where. Am not saying it is better, but after trying to adapt this solution to a challenging problem today, was finding the syntax for where no so intuitive. In the end, not sure whether it would have possible with where, but found the following method lets you have a look at the subset before you modify it and it for me led more quickly to a solution. Works for the OP here of course as well.
You deliberately set a value on a slice of a dataframe as Pandas so often warns you not to.
This answer shows you the correct method to do that.
The following gives you a slice:
df.loc[df['age1'] - df['age2'] > 0]
..which looks like:
age1 age2
0 23 10
1 45 20
Add an extra column to the original dataframe for the values you want to remain after modifying the slice:
df['diff'] = 0
Now modify the slice:
df.loc[df['age1'] - df['age2'] > 0, 'diff'] = df['age1'] - df['age2']
..and the result:
age1 age2 diff
0 23 10 13
1 45 20 25
2 21 50 0
