Michael Deistler

As of now, you can also use [truncated proposals](https://arxiv.org/abs/2210.04815) to avoid the leakage issue: ```python from sbi.inference import SNPE from sbi.utils import get_density_thresholder, RestrictedPrior inference = SNPE(prior) proposal = prior...

Thanks for creating this! Just to re-iterate: this error will occur, e.g., during `.train()` of SNPE-C (second round), when `prior.log_prob()` is called. This gives an error because it is evaluated...

And to reproduce: ```python from torch import ones, zeros from torch.distributions import AffineTransform, TransformedDistribution, Uniform base = Uniform(zeros(1), ones(1)) dist = TransformedDistribution(base, AffineTransform(zeros(1), ones(1))) dist.support.check(100*ones(1)) # -> returns tensor([True]) ```

Good point, agreed!

what exactly should `joint_plot` do?

we have `pairplot(..., subset=[0, 2, 3])`, does this work?

ah, I see your point. I am not 100% convinced we need this, maybe we just leave this to users to implement themselves if really needed?

Feel free to change it, but also needs to update [docs](https://github.com/sbi-dev/sbi/blob/main/docs/mkdocs.yml)

I would also prefer to not change this right now -- users will already get a bunch of warnings when the new version is released (thinning for MCMC, importing posterior_nn)...

Thanks for creating this issue! I can not reproduce the error though. Two questions: 1) Does the following code work for you: ```python import torch from torch import eye, ones,...