I'm seeing many small negative numbers, below is a minimal reproducible example, i.e. -1.77635684e-15 is the small negative number produced.
In [1]: import numpy as np
In [2]: np.cumsum(np.array([11.9, 32.6, -32.6, -11.9]))
Out[2]: array([ 1.19000000e+01, 4.45000000e+01, 1.19000000e+01, -1.77635684e-15])
However, if I reorder the numbers, -1.77635684e-15 will be gone, e.g.
In [3]: np.cumsum(np.array([11.9, 32.6, -11.9, -32.6]))
Out[3]: array([11.9, 44.5, 32.6, 0. ])
I understand it likely has something to do with the floating-point arithmetic issue, but I wonder if someone can explain exactly how -1.77635684e-15 is produced in this example or with a more trivial example if one can be found?
I'm using python-3.8 and my number version is
In [4]: np.__version__
Out[4]: '1.19.1'
