在 pandas 中查询关于.astype() function

Question

I am currently learning from an online course where I was told that for the .astype() function to be used no NaN(null) values must be present. However, while typing a program I was careless and did not check for NaN values and used the astype() function. It was an object before and I converted it to boolean, later realizing I had NaN values. However, no errors were raised and upon calling .info() on the panda object , it returned no null values on the column. Please explain. I have attached images of this behavior.

Answer 1

.astype can be dangerous. I suggest you only use it for str or 'O' conversions. For numerics there are dedicated pd.to_numeric , pd.to_datetime or pd.to_timedelta methods. Sadly, bools don't have an equivalent method.

.astype throws an error if you try to convert something that cannot be converted. Here NaN is a float, which cannot fit into an integer container.

pd.Series(np.NaN).astype(int)
#ValueError: Cannot convert non-finite values (NA or inf) to integer

But then there's bool , and while .astype is doing nothing incorrect, it's probably not doing what you want it to do. The issue is that bool(np.NaN) is perfectly well defined.

bool(np.NaN)
#True

So, .astype has no issues converting np.NaN to True when you use it.

pd.Series([True, np.NaN, False]).astype(bool)
#0     True
#1     True  <- NaN became True. Did you really want that?
#2    False
#dtype: bool 

---

Currently, there is no nullable Bool type, so you can't have a bool dtype with NaN . You either need to use an object column and where after the .astype

s = pd.Series([True, np.NaN, False])
s.astype(bool).astype('O').where(s.notnull())
#0     True
#1      NaN
#2    False
#dtype: object

Or you could try the Int64 dtype

s = pd.Series([True, np.NaN, False])
s.astype(bool).astype('Int64').where(s.notnull())
#0      1
#1    NaN
#2      0
#dtype: Int64