torch_truncnorm

torch_truncnorm copied to clipboard

# Problem One sided truncated normal distributions are not differentiable with respect to their parameters. # Solution Add explicit handling of truncated normal distributions when calculating the probability density and...

BenZickel

Upload to pypi

2

Thanks for making this. Very helpful! It'd be really nice if you could `pip install` it.

oliverdain

Invalid value when calling log_prob after sample

5

My code looks like ```py m = TruncatedNormal(loc, scale, 0, 1) action_pt = m.sample() return m.log_prob(action_pt) ``` It looks like `action_pt` can take the value `1.0` and causes `log_prob` to...

louisabraham

Dear Anton, I'm currently trying to add the expand function to the truncated normal class you provided, as I need to generate multiple truncated normal distributions in parallel. I've made...

XiankangTang