Jan

docs: move general class docstrings from __init__ to class level for better API refs

3

On the new docs website we now have nice API references, however, the docstrings with a high-level description of a class do not show up because there are placed in...

bug: docs building is failing because of missing score posterior

3

in #1497 we exchanged the score posterior for the vector field posterior. We need to update the `docs` accordingly to fix this issue: 

bug: pymc nuts failing with ValueError

1

when using pymc NUTS with python 3.10 and pymc > 5.20.0 we get: ``` Traceback (most recent call last): File "/Users/janteusen/.local/share/uv/python/cpython-3.11.11-macos-aarch64-none/lib/python3.11/threading.py", line 1045, in _bootstrap_inner self.run() File "/Users/janteusen/qode/sbi/.venv/lib/python3.11/site-packages/rich/live.py", line 32,...

sampling via pymc raises TypeError since pymc v5.20.1

11

see #1390 for details. To fix this issue, we should check whether `pymc` made a new release and fixed the problem and then remove the `fail` mark from the tests...

improve interface for custom density estimators passed to inference classes

1

In all inference methods (child classes of `NeuralInference`) we generally allow to either pass a string for building the density estimator internally, e.g., `NPE(density_estimator="maf")` or we allow passing a plain...

tests: switch to a bi-modal posterior for testing

3

We are testing our inference methods using a linear Gaussian simulator with a Gaussian target posterior. Should be switch to a bimodal posterior instead to detect possible shortcomings of inference...

docs: add tutorial / how-to-guide for embedding time-series data, e.g., using causalCNNs

6

We now have a bunch of new embedding nets to process sequential data, e.g., the CausalCNN or the TransformerEmbedding (see also #1499 #1494 #1512 ) It would be great to...

docs: add a tutorial / how-to-guide for using pyro with sbi

3

We now can use sbi estimators as pyro models (#1491 ) and it would be nice to have a short guide how to do this. This could be integrated into...

It would be nice to have a short guide on how to tune hyperparameters, e.g., neural architectures in `sbi`. E.g., by using a test set of ~1000 `(theta, x)`s and...

docs: a how-to-guide for using unconditional flows to detect model misspecification

2

To detect a potential misspecification / out of distribution problem of a real-world `x_o`, one could train an unconditional flow on the training data `x` to learn `q(x)`. Given a...