What is the 'fillna()' euiqvalent for dtype 'Int32'?-CodePudding

Short question: How can I set all values that are <1 or <NA> to 1?

Long question: Say I have a pure-int (int32!) pandas column, I used can do this to cap the minimum:

>>> shots = pd.DataFrame([2, 0, 1], index=['foo', 'bar', 'baz'], columns={'shots'}, dtype='int32')
shots
     shots
foo      2
bar      0
baz      1

>>> max(shots.loc['foo', 'shots'], 1)
2

>>> max(shots.loc['bar', 'shots'], 1)
1

So far, so good. Now, say the dtype of column shots changes from 'int32' to Int32, allowing <NA>. This gets me in trouble when accessing <NA> records. I get this error:

>>> shots = pd.DataFrame([2, np.nan, 1], index=['foo', 'bar', 'baz'], columns={'shots'}, dtype='Int32')
     shots
foo      2
bar   <NA>
baz      1

>>> max(shots.loc['bar', 'shots'], 1)    
`TypeError: boolean value of NA is ambiguous`

What should I do?

My first intuition was to say "Ok, let's fill values, then apply max().". But that also fails:

>>> shots.loc[idx, 'shots'].fillna(1)

AttributeError: 'NAType' object has no attribute 'fillna'

--> What is the most pandiastic/pydantic way to apply a condition to <NA> values, i.e., setting all <NA> to 1, or applying some other form of basic match, such as max(<NA>, 1)?

Versions

Python 3.8.6
Pandas 1.2.3
Numpy 1.19.2

CodePudding user response：

idx should be a collection else if it's a scalar you get a scalar value:

# idx = 'bar'

>>> shots.loc[idx, 'shots']
<NA>

>>> shots.loc[idx, 'shots'].fillna(1)
...
AttributeError: 'NAType' object has no attribute 'fillna'

>>> shots.loc[[idx], 'shots'].fillna(1)
bar    1
Name: shots, dtype: Int32

The question is how idx is defined?

Old answer

Your problem is not reproducible for me.

shots = pd.DataFrame({'shots': [2, 1, pd.NA]}, dtype=pd.Int32Dtype())
idx = [2]

>>> shots
   shots
0      2
1      1
2   <NA>

>>> shots.dtypes
shots    Int32
dtype: object

>>> shots.loc[idx, 'shots'].fillna(1)
2    1
Name: shots, dtype: Int32

Versions:

Python 3.9.7
Pandas 1.4.1
Numpy 1.21.5