normal_lcdf and normal_lccdf give infinite gradients for infinite inputs.
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 839
- Forks
- 220
- Avg merge
- 2d 4h
- Merged PRs (30d)
- 14
Description
Description
Consider a program:
data{
real y;
}
parameters{
real mu;
real<lower = 0> sigma;
}
model{
target += normal_lcdf(y | mu, sigma);
}
When y is infinite (and given that mu and sigma are finite, here ensured by declaring them as parameters), then the gradient is zero. However, it seems that Stan yields infinite gradients here.
This came up "in the wild" here:
https://discourse.mc-stan.org/t/conditional-truncation-including-inf/30627
If possible, it would be nice to special-case infinite y and return the correct gradients. However, there's some question beyond my expertise of whether this special-casing would play nicely with vectorization in opencl.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by locating the normal_lcdf and normal_lccdf entry points and tracing their autodiff behavior for infinite inputs. Compare the scalar, vectorized, and OpenCL paths mentioned in the issue, then find the relevant existing tests. Done means infinite inputs produce zero gradients while finite-input behavior remains unchanged.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- backend
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100