Fritz Obermeyer

Results 122 issues of Fritz Obermeyer

This is a long term issue that probably won't be fixed until the [next generation](https://github.com/pyro-ppl/funsor) of distribution libraries. It would be useful if distributions could be sliced along their `sample_shape...

enhancement
discussion

@eb8680 proposed the idea of making a compatibility shim for `pyro.contrib.funsor` so models could be written in a backend agnostic way. (this task is being migrated out of #2968)

enhancement

This attempts to apply `LocScaleReparam` in `AutoReparam` only for latent variables whose prior distributions depends on upstream variables. In case the prior has fixed parameters, there is no need to...

enhancement
WIP
Blocked

This issue tracks new errors in the PyTorch 1.10 jit, e.g. ```sh pytest tests/infer/test_jit.py::test_dirichlet_bernoulli -k Jit -vx --runxfail ``` ``` __ test_dirichlet_bernoulli[JitTraceEnum_ELBO-False] __ Elbo = , vectorized = False @pytest.mark.parametrize("vectorized",...

bug
help wanted

A common pattern in applied probabilistic modeling is to implement domain-specific distributions, constraints, and transforms for individual data science projects. That is, it is common for users to extend Pyro/NumPyro...

help wanted
Examples

Currently guides using `safe_log()` cannot be serialized due to https://github.com/pytorch/pytorch/issues/22329. `safe_log()` is used in our log backend for einsum internally, as used in variable elimination and funsor contraction. One option...

enhancement

Consider the model ```py def model1(data): # -x is exponentially distributed x = pyro.sample("x", TransformedDistribution(Uniform(0, 1), ExpTransform().inv) pyro.sample("obs", Normal(0, -x), obs=data) ``` and its reparametrized version ```py model2 = poutine.reparam(model1,...

question
discussion

This issue proposes a workflow speed up HMC in high dimensions: first train a structured Gaussian guide via SVI (which is cheap), then use the structured sqrt(covariance) as a preconditioner...

enhancement

@cwhelan writes > I have a very newb question about developing models in pyro: are there any tips and tricks that make debugging problems in your model easier? I'm new...

documentation
warnings & errors