def test_decorator(self):
"""When applied to a function it allows it to work with scalars."""
# Setup
function = MagicMock()
function.__doc__ = 'Docstring of the original function.'
function.return_value = np.array(['return_value'])
instance = MagicMock()
args = ['positional', 'arguments']
kwargs = {'keyword': 'arguments'}
expected_result = 'return_value'
# Run (Decorator)
scalarized_function = scalarize(function)
# Check (Decorator)
assert callable(scalarized_function)
assert scalarized_function.__doc__ == 'Docstring of the original function.'
# Run (Decorated function)
result = scalarized_function(instance, 0, *args, **kwargs)
# Check (Decorated function)
assert result == expected_result
function.assert_called_once_with(instance, np.array([0]), *args,
**kwargs)
instance.assert_not_called()
assert instance.method_calls == []
def _brentq_cdf(self, value):
"""Helper function to compute percent_point.
As scipy.stats.gaussian_kde doesn't provide this functionality out of the box we need
to make a numerical approach:
- First we scalarize and bound cumulative_distribution.
- Then we define a function `f(x) = cdf(x) - value`, where value is the given argument.
- As value will be called from ppf we can assume value = cdf(z) for some z that is the
value we are searching for. Therefore the zeros of the function will be x such that:
cdf(x) - cdf(z) = 0 => (becasue cdf is monotonous and continous) x = z
Args:
value (float):
cdf value, that is, in [0,1]
Returns:
callable:
function whose zero is the ppf of value.
"""
# The decorator expects an instance method, but usually are decorated before being bounded
bound_cdf = partial(scalarize(GaussianKDE.cumulative_distribution),
self)
def f(x):
return bound_cdf(x) - value
return f