_minimize_scalar_bounded: reference before assignment

[MartinGrignard]

Hi,
I've recently changed my docker configuration and, while using chaospy package, I came across an error which never occurred before.
It looks like it comes from the fact that fu is not instantiated before the while loop but I'm not sure.
I hope this report will help.

Reproducing code example:

import chaospy as cp

distribution = cp.J(cp.Uniform(0, 1), cp.Uniform(0, 1))
abscissas, weights = cp.generate_quadrature(7, distribution, rule="Leja", sparse=True)

Error message:

This error message occured in a Jupyter notebook:

---------------------------------------------------------------------------
UnboundLocalError                         Traceback (most recent call last)
<ipython-input-1-b2f64640c4d5> in <module>
      1 import chaospy as cp
      2 dist=cp.J(cp.Uniform(0, 1), cp.Uniform(0, 1))
----> 3 points = cp.generate_quadrature(2, dist, rule="Leja", sparse=True)

/usr/local/lib/python3.7/dist-packages/chaospy/quadrature/frontend.py in generate_quadrature(order, dist, rule, sparse, accuracy, growth, recurrence_algorithm)
    134         from . import sparse_grid
    135         return sparse_grid.construct_sparse_grid(
--> 136             order, dist, rule=rule, accuracy=accuracy, growth=growth)
    137 
    138     if not isinstance(rule, str):

/usr/local/lib/python3.7/dist-packages/chaospy/quadrature/sparse_grid.py in construct_sparse_grid(order, dist, rule, accuracy, growth, recurrence_algorithm)
    122 
    123     x_lookup, w_lookup = _construct_lookup(
--> 124         orders, dist, rule, accuracy, growth, recurrence_algorithm)
    125     collection = _construct_collection(
    126         order, dist, x_lookup, w_lookup)

/usr/local/lib/python3.7/dist-packages/chaospy/quadrature/sparse_grid.py in _construct_lookup(orders, dists, rules, accuracy, growth, recurrence_algorithm)
    182                 rule=rule,
    183                 growth=growth,
--> 184                 recurrence_algorithm=recurrence_algorithm,
    185             )
    186             x_lookup[-1].append(abscissas)

/usr/local/lib/python3.7/dist-packages/chaospy/quadrature/frontend.py in generate_quadrature(order, dist, rule, sparse, accuracy, growth, recurrence_algorithm)
    157 
    158     quad_function = QUAD_FUNCTIONS[rule]
--> 159     abscissas, weights = quad_function(order, dist, **kwargs)
    160 
    161     assert len(weights) == abscissas.shape[1]

/usr/local/lib/python3.7/dist-packages/chaospy/quadrature/leja.py in quad_leja(order, dist, rule, accuracy, recurrence_algorithm)
    104             *[fminbound(
    105                 objective, abscissas[idx], abscissas[idx+1], full_output=1)[:2]
--> 106               for idx in range(len(abscissas)-1)]
    107         )
    108         index = numpy.argmin(vals)

/usr/local/lib/python3.7/dist-packages/chaospy/quadrature/leja.py in <listcomp>(.0)
    104             *[fminbound(
    105                 objective, abscissas[idx], abscissas[idx+1], full_output=1)[:2]
--> 106               for idx in range(len(abscissas)-1)]
    107         )
    108         index = numpy.argmin(vals)

/usr/local/lib/python3.7/dist-packages/scipy/optimize/optimize.py in fminbound(func, x1, x2, args, xtol, maxfun, full_output, disp)
   1734                'disp': disp}
   1735 
-> 1736     res = _minimize_scalar_bounded(func, (x1, x2), args, **options)
   1737     if full_output:
   1738         return res['x'], res['fun'], res['status'], res['nfev']

/usr/local/lib/python3.7/dist-packages/scipy/optimize/optimize.py in _minimize_scalar_bounded(func, bounds, args, xatol, maxiter, disp, **unknown_options)
   1868             break
   1869 
-> 1870     if np.isnan(xf) or np.isnan(fx) or np.isnan(fu):
   1871         flag = 2
   1872 

UnboundLocalError: local variable 'fu' referenced before assignment

Scipy/Numpy/Python version information:

1.3.3 1.17.4 sys.version_info(major=3, minor=7, micro=3, releaselevel='final', serial=0)

[andyfaff]

Thank you for this report. As you say fu is not initialised before the while loop, and the while loop doesn't run in this instance because it's within some kind of tolerance.
It looks to me like fu could be initialised as fu = np.inf without an issue, but I'm not familiar with this optimizer. It looks like it could be fixed relatively simply, would you be interested in making a PR?

[MartinGrignard]

Thanks for this quick answer!
I can try your solution. If it works I'll do a PR.

[WarrenWeckesser]

Here's a simple example to reproduce the problem:

In [1]: from scipy.optimize import fminbound

In [2]: def fun(x):
   ...:     return x**2
   ...:     

In [3]: fminbound(fun, 0, 0)
---------------------------------------------------------------------------
UnboundLocalError                         Traceback (most recent call last)
<ipython-input-3-178a9623aadc> in <module>
----> 1 fminbound(fun, 0, 0)

~/mc37scipy/lib/python3.7/site-packages/scipy-1.5.0.dev0+0a7b39a-py3.7-macosx-10.7-x86_64.egg/scipy/optimize/optimize.py in fminbound(func, x1, x2, args, xtol, maxfun, full_output, disp)
   1744                'disp': disp}
   1745 
-> 1746     res = _minimize_scalar_bounded(func, (x1, x2), args, **options)
   1747     if full_output:
   1748         return res['x'], res['fun'], res['status'], res['nfev']

~/mc37scipy/lib/python3.7/site-packages/scipy-1.5.0.dev0+0a7b39a-py3.7-macosx-10.7-x86_64.egg/scipy/optimize/optimize.py in _minimize_scalar_bounded(func, bounds, args, xatol, maxiter, disp, **unknown_options)
   1878             break
   1879 
-> 1880     if np.isnan(xf) or np.isnan(fx) or np.isnan(fu):
   1881         flag = 2
   1882 

UnboundLocalError: local variable 'fu' referenced before assignment

[andyfaff]

@MartinGrignard you can use that easier example as a test to make sure it doesn't happen again.

Normally when I add a test I call the test something like test_gh11207 so that we know why the test was made.

[MartinGrignard]

@andyfaff it looks like it passed all the tests 👍