ENH : remove an expected warning in BarycentricInterpolator

[tlunet]

Reference issue

See this comment on an other project.

What does this implement/fix?

It remove an expected warning when evaluating a BarycentricInterpolator object (see here...).

The reason why : in the evaluation formula for the barycentric interpolation (see https://parallel-in-time.org/pySDC/pySDC/core.html#module-core.Lagrange), there is a division by zero when an evaluated point coincides with a node. It can be reproduced with the following code :

from scipy.interpolate import BarycentricInterpolator
interp = BarycentricInterpolator([0, 1], [1, 2])
yi = interp(interp.xi)

which outputs :

/prog/anaconda/lib/python3.8/site-packages/scipy/interpolate/_polyint.py:658: RuntimeWarning: divide by zero encountered in true_divide
  p = np.dot(c,self.yi)/np.sum(c,axis=-1)[...,np.newaxis]

Since this division by zero is actually expected and the values corrected later, this PR wraps the line raising the warning with a with np.errstate(divide='ignore'): statement.

Additional information

Spyder IDE specific files added to .gitignore

[ev-br]

Looks reasonable, thanks @tlunet !

Would be nice to add test in a follow-up PR: use just the reproducer from the OP, with warnings turned to errors.

[tlunet]

Looks reasonable, thanks @tlunet !

Would be nice to add test in a follow-up PR: use just the reproducer from the OP, with warnings turned to errors.

Sure, I can do that. You may just have to explain me what the "reproducer from the OP" is 😅

[ev-br]

Meant the three-line code snippet after It can be reproduced with the following code in your original PR message :-).