Implement ICDF Methods for Truncated Distributions #1938
Conversation
|
Fixed an issue where the icdf functions would return non-NaN values for inputs outside of the range [0,1], although the values in the range are correct. Now the icdf function returns a NaN for inputs outside the range [0,1], which makes it consistent with other distributions as well. |
|
All formatting and relevant distribution tests pass! Ready for review. |
There was a problem hiding this comment.
Thanks @TheSkyentist Could you double check if our tests already cover the change?
|
|
||
| def icdf(self, q): | ||
| ppf = self.base_dist.icdf(q * self._cdf_at_high) | ||
| return jnp.where(q > 1, jnp.nan, ppf) |
There was a problem hiding this comment.
why do we need to cover q > 1?
| ) | ||
| return jnp.where(jnp.logical_or(q < 0, q > 1), jnp.nan, ppf) |
There was a problem hiding this comment.
like above, why this where is needed?
There was a problem hiding this comment.
See above comment :-)
|
I think the tests cover the change at least as far as I can see, but perhaps I'm not sure what you mean, could you clarify slightly? Thanks :-) |
Implementation of generic Inverse Cumulative Distribution Function (ICDF) methods for Truncated Distributions.
Sampling method for Truncated Distribution already calculates the ICDF, so the initial commit is refactoring the code into its own method. Still needs to be tested so starting this as a draft.
Motivated by #1937 for use with Nested Sampling with
jaxns.