Unexpected side effect using numpy trig functions

[josiahslack]

I ran across this yesterday.

I type the following into a python REPL:

`
import numpy as np
import pint

ureg = pint.UnitRegistry()
Q_ = ureg.Quantity
theta = Q_(30, ureg.degrees)
type(theta.m) # returns int
s_theta = np.sin(theta)
type(theta.m) # returns numpy.ndarray
`
Printing theta, using comparisons, etc. doesn't show that anything has changed about it. I ran into trouble when I was saving an angle's magnitude to a JSON file, and json.dump complained that it didn't know how to write 'array(0.0)'. I worked backwards, and eventually isolated the point where the type of the angle's magnitude changed.

[hgrecco]

Thanks for reporting. We should not transform scalars to arrays.
And it is not numpy's fault

>>> type(np.sin(.3))
<class 'numpy.float64'>
>>> type(.3)
<class 'float'>

[CD3]

I ran into this too.

import pint
import numpy

ureg = pint.UnitRegistry()

Angle = ureg.Quantity(30,'degree')

Cos = ureg.Quantity(numpy.sqrt(3.)/2.)
print Angle, type(Angle), Angle.magnitude, type(Angle.magnitude)
print Cos, type(Cos), Cos.magnitude, type(Cos.magnitude)
print "cos of {Angle:.1f} is {Cos:.1e}".format(Angle=Angle, Cos=Cos)

Cos = numpy.cos(Angle)
print Angle, type(Angle), Angle.magnitude, type(Angle.magnitude)
print Cos, type(Cos), Cos.magnitude, type(Cos.magnitude)
print "cos of {Angle:.1f} is {Cos:.1e}".format(Angle=Angle, Cos=Cos)

This throws an error on the second format() call.

30 degree <class 'pint.unit.Quantity'> 30 <type 'int'>
0.866025403784 dimensionless <class 'pint.unit.Quantity'> 0.866025403784 <type 'numpy.float64'>
cos of 30.0 degree is 8.7e-01 dimensionless
30 degree <class 'pint.unit.Quantity'> 30 <type 'numpy.ndarray'>
0.866025403784 dimensionless <class 'pint.unit.Quantity'> 0.866025403784 <type 'numpy.float64'>
Traceback (most recent call last):
  File "./test.py", line 18, in <module>
    print "cos of {Angle:.1f} is {Cos:.1e}".format(Angle=Angle, Cos=Cos)
  File "/usr/lib64/python2.7/site-packages/pint/quantity.py", line 141, in __format__
    format(obj.magnitude, remove_custom_flags(spec)),
ValueError: Unknown format code 'f' for object of type 'unicode'

It took me a while to figure out what was going on. The call to numpy.cos changes the type of Angle. It isn't just trig functions either, numpy.abs will also change the type of its argument.

I would expect any quantities that get passed into the numpy functions are left unmodified. Is pint modifying the argument intentionally for some reason? If not, could the numpy hooks just make a copy of the arguments to pass to the numpy functions?

[CD3]

A workaround is to pass copies of the quantity to the numpy functions using the copy module

import pint
import numpy
from copy import *

ureg = pint.UnitRegistry()

Angle = ureg.Quantity(30,'degree')

Cos = ureg.Quantity(numpy.sqrt(3.)/2.)
print Angle, type(Angle), Angle.magnitude, type(Angle.magnitude)
print Cos, type(Cos), Cos.magnitude, type(Cos.magnitude)
print "cos of {Angle:.1f} is {Cos:.1e}".format(Angle=Angle, Cos=Cos)

Cos = numpy.cos(copy(Angle)) # pass copy of quantity so it won't be modified.
print Angle, type(Angle), Angle.magnitude, type(Angle.magnitude)
print Cos, type(Cos), Cos.magnitude, type(Cos.magnitude)
print "cos of {Angle:.1f} is {Cos:.1e}".format(Angle=Angle, Cos=Cos)

works as expected

30 degree <class 'pint.unit.Quantity'> 30 <type 'int'>
0.866025403784 dimensionless <class 'pint.unit.Quantity'> 0.866025403784 <type 'numpy.float64'>
cos of 30.0 degree is 8.7e-01 dimensionless
30 degree <class 'pint.unit.Quantity'> 30 <type 'int'>
0.866025403784 dimensionless <class 'pint.unit.Quantity'> 0.866025403784 <type 'numpy.float64'>
cos of 30.0 degree is 8.7e-01 dimensionless

[hgrecco]

Review after #905

[hgrecco]

Solved by #905