manuelgloeckler

Replace redundant `x_shape`

1

## Description The new `DensityEstimators` will have a `_condition_shape` attribute, which is required to correctly handle batch shapes on `x_o`. Currently, all posterior classes also have a `_x_shape` attribute which...

Add `sample_batched` and `log_prob_batched` to posteriors

1

## Description The current posterior estimators only support a single x_o (or iid x_os), for sampling and log_prob. This can be inefficient when we want to sample from a **batch**...

Add a common MDN network interface (consistent with new DensityEstimators)

2

## Description The MDN (Mixture Density Network) — a Gaussian Mixture Density network — currently receives several "special" treatments in the codebase: - **SNPE_A** relies on it and implements a...

We currently only have the `DensityEstimator` abstraction, but we might want to expand this for other approaches. I would suggest something like this: ```python class Estimator(nn.Module): def loss(...) class DensityEstimator(Estimator)...

Integrate new density estimator interface into SBI

5

## Description Following #952 we need to integrate the new Density Estimator class into SBI ## Checklist Following steps: - [ ] 1) Change `posterior_nn` helper function to return the...

Prepare SBI for diffusion models

2

**Set up SBI to incorporate diffusion models and co** Set up the integration of diffusion models and score-based approaches. - [ ] Common sdes - [ ] Transformers - [...

Base estimator class

1

## What does this implement/fix? Explain your changes Implements an abstract base class for estimators, which has: - `input_shape` - `condition_shape` - abstract loss function. ## Does this close any...

feat: batched sampling for MCMC

3

## What does this implement/fix? Explain your changes This pull request aims to implement the `sample_batched` method for MCMC. ## Current problem - `BasePotential` can either "allow_iid" or not. Hence,...

Slow L-C2ST test fails locally

1

**Describe the bug** L-C2ST (slow) tests fail locally. ```bash FAILED tests/lc2st_test.py::test_lc2st_true_positiv_rate[LC2ST_NF] - AssertionError: LC2ST p-values too small, test should be accepted at least 95.0% of the time, but was accepted...

Fix pickle issues in MCMC posterior + test

1

## What does this implement/fix? Explain your changes This PR address the issue of the MCMCPosterior not being pickable after running sample. ## Does this close any currently open issues?...