The inf bounds for truncated normal is inconsistent

[damar-wicaksono]

The lower or upper bound of a truncated normal distribution may be set to either -inf or inf, but this is inconsistent with the implementation of the normal distribution. While the normal distribution is strictly speaking unbounded, in UQTestFuns the bounds are set such that the difference between 1.0 and the probability contained of the distribution between the two bounds are less than 1e-15.
This means that the bounds are set to:

lb = mu - 8.22 * sigma
ub = mu + 8.22 * sigma

The truncated normal may also be set to have infinite bounds which is the same as the untruncated normal distribution. In that case, for consistency, the bounds should take the bounds set for the normal distribution.

my_input_1 = uqtestfuns.UnivariateInput(distribution="trunc-normal", parameters=[0, 1, -np.inf, np.inf])  # a standard normal
my_input_2 = uqtestfuns.UnivariateInput(distribution="normal", parameters=[0, 1])  # also a standard normal
assert my_input_1.icdf(1.0) == my_input_2.icdf(1.0)  # But this will fail

[damar-wicaksono]

This issue has been resolved by PR #91.