Benjamin Kurt Miller
Benjamin Kurt Miller
it is possible to use custom datasets (like [this](https://pytorch.org/tutorials/beginner/basics/data_tutorial.html#creating-a-custom-dataset-for-your-files)) as long as the types conform with `swyft.Dataset`. The prior must be properly specified in this case. A tutorial would help...
It would be nice if it was possible to easily combine truncation bounds. Right now truncation happens over every estimated dimension of a `MarginalPosterior`... The process could be made a...
Evaluating the icdf and taking numerical derivatives doesn't work accurately, even with relative tolerance of 1e-4, for simple normal distributions (and not even some uniform distributions). run the `pytest tests/prior_test.py`...
EDITED: DirectoryStore is slow, presumably due to saving to disk. However, MemoryStore is also really slow using some parameters. (namely when you use it for a dataset which does not...
It would be great to also create a "to pytorch dataset" feature where a zarr store would produce a dataset which does not access the zarr store. Rather it copies...
Our repo is needlessly large because of the excessive saves of notebooks from previous git commits. This can be undone. We should do so. A method can be found here:...
For reproducibility there is the option to split the seed, we should use it. https://numpy.org/doc/stable/reference/random/index.html? https://github.com/undark-lab/swyft/blob/07a5bffb543b4afab050cfc63c897215ccb6b355/swyft/store/store.py#L94-L131
I propose that we operate under the assumption that all of our users will be able to instantiate the class again, then load the parameters using state_dict, as usual in...
make ensemble predictions simple, in an effort to produce more conservative posterior approimations
This paper by Gilles and co, [Averting A Crisis In Simulation-Based Inference](https://arxiv.org/abs/2110.06581), advocates using ensembles to combat non-conservative posterior estimates. I think it makes sense to allow for simple ensemble...
This is done within dynesty in the following manner: https://dynesty.readthedocs.io/en/latest/quickstart.html#example-conditional-priors It's a bit more complicated for us to do this because we have to consider marginalizing over them.