I am particularly talking about Pandas version 0.11 as I am busy replacing my uses of .ix with either .loc or .iloc. I like the fact that differentiating between .loc and .iloc communicates whether I am intending to index by label or integer position. I see that either one will accept a boolean array as well but I would like to keep their usage pure to clearly communicate my intent.
In 11.0 all three methods work, the way suggested in the docs is simply to use df[mask]
. However, this is not done on position, but purely using labels, so in my opinion loc
best describes what's actually going on.
Update: I asked on github about this, the conclusion being that df.iloc[msk]
will give a NotImplementedError
(if integer indexed mask) or ValueError
(if non-integer indexed) in pandas 11.1
.
In [1]: df = pd.DataFrame(range(5), list('ABCDE'), columns=['a'])In [2]: mask = (df.a%2 == 0)In [3]: mask
Out[3]:
A True
B False
C True
D False
E True
Name: a, dtype: boolIn [4]: df[mask]
Out[4]:a
A 0
C 2
E 4In [5]: df.loc[mask]
Out[5]:a
A 0
C 2
E 4In [6]: df.iloc[mask] # Due to this question, this will give a ValueError (in 11.1)
Out[6]:a
A 0
C 2
E 4
Perhaps worth noting that if you gave mask integer index it would throw an error:
mask.index = range(5)
df.iloc[mask] # or any of the others
IndexingError: Unalignable boolean Series key provided
This demonstrates that iloc isn't actually implemented, it uses label, hence why 11.1 will throw a NotImplementedError
when we try this.