There seems to be a subtle difference between numpy.float and numpy.float64.
>>> import numpy as np
>>> isinstance(2.0, np.float)
True
>>> isinstance(2.0, np.float64)
False
Can someone clarify this? Thanks
2 Answers
np.float is an alias for python float type.
np.float32 and np.float64 are numpy specific 32 and 64-bit float types.
float?
Init signature: float(self, /, *args, **kwargs)
Docstring:
float(x) -> floating point number
Convert a string or number to a floating point number, if possible.
Type: type
np.float?
Init signature: np.float(self, /, *args, **kwargs)
Docstring:
float(x) -> floating point number
Convert a string or number to a floating point number, if possible.
Type: type
np.float32?
Init signature: np.float32(self, /, *args, **kwargs)
Docstring: 32-bit floating-point number. Character code 'f'. C float compatible.
File: c:\python\lib\site-packages\numpy\__init__.py
Type: type
np.float64?
Init signature: np.float64(self, /, *args, **kwargs)
Docstring: 64-bit floating-point number. Character code 'd'. Python float compatible.
File: c:\python\lib\site-packages\numpy\__init__.py
Type: type
Thus, when you do isinstance(2.0, np.float), it is equivalent to isinstance(2.0, float) as 2.0 is a plain python built-in float type... and not the numpy type.
isinstance(np.float64(2.0), np.float64) would obviously be True.
1 Comment
Edison
I had an arrival delay time that was
float64. When I converted the departure delay time type from int64 to match the arrival delay type, I didn't catch that I used fillna().astype('float') and the data/scatter plot changed a lot. I couldn't figure out what was going on. When I corrected it to fillna().astype('float64') all was good. Not sure why it made such a difference when all the values looked like 55.00, 130.00 etc.This would seem to be the difference between a 32-bit floating point number and a 64-bit floating point number (in C, a float vs. a double), and 2.0 ends up being a 32-bit floating point number.
1 Comment
zell
Sorry, I have to downvote: Your answer seems wrong. Here is what I just tried: isinstance(1e200, np.float64), which gives false. In addition, isinstance(2.0, np.float32) gives false too.
numpy.floatis Python's float class. Do you meannumpy.float_with an underscore at the end? That is the same asnumpy.float64.