I'm having a problem with np.round, np.around where it is not rounding properly. I can't include code, because when I do it manually set the value (as opposed to use the my data), the return works, but here is the output:
In [177]: a
Out[177]: 0.0099999998
In [178]: np.round(a,2)
Out[178]: 0.0099999998
In [179]: np.round(a,1)
Out[179]: 0.0
What am I missing? The dtype of a is float32, do I need to change this?
Try creating np.float32(0.01) and you will see your answer. You are getting the precision you can already.
>>> import numpy as np
>>> x = 0.01
>>> epsilon = 0.01 - np.float32(0.01)
>>> for n in np.arange(x - 10*epsilon, x + 10*epsilon, epsilon):
... print(repr(np.float32(n)))
...
0.0099999979
0.0099999979
0.0099999979
0.0099999988
0.0099999988
0.0099999988
0.0099999988
0.0099999998
0.0099999998
0.0099999998
0.0099999998
0.0099999998
0.010000001
0.010000001
0.010000001
0.010000001
0.010000002
0.010000002
0.010000002
0.010000002
np.float64 .. but note that it doesn't avert the "problem", it only makes the error smaller.
x = numpy.round(0.0099999999, decimals=2) use float(x) to convert to 64-bit precision, or '{0}'.format(x, precision=2) if you need to write the number to two decimal places.Note there seems to be an issue with python's round function and numpy.float64 types. See example below:
In [88]: round(np.float64(16.259766999999947), 4)
Out[88]: 16.259799999999998
The only way I could fix this is to convert the numpy.float64 to a float before using the round function as below:
In [89]: round(float(np.float64(16.259766999999947)), 4)
Out[89]: 16.2598
x = 5672.1001, then np.round(float(np.float64(x)),1) yields 5672.1000000000004.It does not work for me in cases the number is python's float or np.float64.
once I change the number to np.float32(), it works!
type(85.10000000000001), np.round(85.10000000000001,decimals=2), np.round(np.float32(85.10000000000001),decimals=2)
Results: (float, 85.10000000000001, 85.1)
