minimize_ipopt raises TypeError for `tol` argument

[timmens]

Problem

Minimization using cyipopt.minimize_ipopt raises a TypeError for the tol argument on the same code in version 1.3.0 but not in version 1.2.0. Please take a look at the minimal reproducible example below.

Minimal reproducible example

import cyipopt
import numpy as np

def fun(x):
    return np.sum(x ** 2)

res = cyipopt.minimize_ipopt(
    fun=fun,
    x0=np.zeros(2),
    tol=1e-9,
)

Error excerpt:

TypeError: Invalid option for IPOPT: b'tol': 1e-09 (Original message: "Invalid option type")

Probably caused by

It seems that cyipopt.minimize_ipopt calls the function cyipopt.scipy_interface._minimize_ipopt_iv internally (l.534), which converts the argument tol to a numpy.float (l.668), which then causes nlp.add_option(option, value) (l.598) to raise the TypeError.

---

Full error message

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
File ~/miniforge3/envs/[...]/lib/python3.10/site-packages/cyipopt/scipy_interface.py:599, in minimize_ipopt(fun, x0, args, kwargs, method, jac, hess, hessp, bounds, constraints, tol, callback, options)
    598 try:
--> 599     nlp.add_option(option, value)
    600 except TypeError as e:

File ~/miniforge3/envs/[...]/lib/python3.10/site-packages/cyipopt/cython/ipopt_wrapper.pyx:503, in ipopt_wrapper.Problem.add_option()

TypeError: Invalid option type

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
Cell In[2], line 12
      7     return np.sum(x ** 2)
      9 x0 = np.zeros(2)
---> 12 res = minimize_ipopt(
     13     fun=fun,
     14     x0=x0,
     15     tol=1e-9,
     16 )

File ~/miniforge3/envs/[...]/lib/python3.10/site-packages/cyipopt/scipy_interface.py:602, in minimize_ipopt(fun, x0, args, kwargs, method, jac, hess, hessp, bounds, constraints, tol, callback, options)
    600     except TypeError as e:
    601         msg = 'Invalid option for IPOPT: {0}: {1} (Original message: "{2}")'
--> 602         raise TypeError(msg.format(option, value, e))
    604 x, info = nlp.solve(x0)
    606 return OptimizeResult(x=x,
    607                       success=info['status'] == 0,
    608                       status=info['status'],
   (...)
    613                       njev=problem.njev,
    614                       nit=problem.nit)

TypeError: Invalid option for IPOPT: b'tol': 1e-09 (Original message: "Invalid option type")

[moorepants]

@mdhaber this may be from some changes you've added in 1.3.0. Looks like tol is converted to a numpy array.

    if tol is not None:
        tol = np.asarray(tol)[()]
        if tol.ndim != 0 or not np.issubdtype(tol.dtype, np.number) or tol <= 0:
            raise ValueError('`tol` must be a positive scalar.')

[moorepants]

I've made a simple fix for this and can make a bug fix release soon.

[mdhaber]

Hmm. If tol is a Python float, tol = np.asarray(tol)[()] makes it NumPy float, which is a Python float.

import numpy as np
x = 1.0
y = np.asarray(x)[()]
isinstance(y, type(x))  # True

So I didn't think that was backward incompatible. I added a test of tol for SciPy's methods to make sure it was being passed to SciPy correctly (https://github.com/mechmotum/cyipopt/blob/0b6e7494a1cafc19f0dd3b683ace7bad89e28b3c/cyipopt/tests/unit/test_scipy_optional.py#L344), but I guess there were no tests of the tol parameter before that checked it for IPOPT.

In any case, sorry for the trouble.

[moorepants]

No trouble, its part of the process :)