Code for the manuscript

Fast and Reliable Probabilistic Reflectometry Inversion with Prior-Amortized Neural Posterior Estimation

by Vladimir Starostin¹, Maximilian Dax², Alexander Gerlach³, Alexander Hinderhofer³, Álvaro Tejero-Cantero¹, and Frank Schreiber³.

• [1] University of Tübingen, Tübingen, Germany
• [2] Max Planck Institute for Intelligent Systems Tübingen, Germany
• [3] Institute of Applied Physics, University of Tübingen, Tübingen, Germany.

Accepted for publication in Science Advances.

To install the package, clone the repository and run the following command in the root directory:

pip install .

The following dependencies are required and will be installed automatically:

• numpy>=1.26.0
• scipy>=1.11.3
• torch>=2.1.0
• nflows>=0.14
• tqdm
• PyYAML
• click
• pytest
• matplotlib
• h5py

To test the installation, run the following command in the root directory:

pytest

This will execute all tests, including doctests in the panpe package and tests in the tests/functional directory.

To run inference, please refer to the notebooks directory for examples of how to use the package for inference.

# imports
from panpe import InferenceModel, ExpDataset, ROOT_DIR

# load the model
model = InferenceModel.from_config("panpe-2layers-xrr")

# load the experimental dataset (three datasets are concatenated together)
exp_dset = ExpDataset(ROOT_DIR / "data/xrr_data.h5", device="cpu")

# run inference
res = model(exp_dset[200])

# plot the results
res.plot_sampled_profiles(show_prior=True)

The panpe-2layers-xrr model for XRR data is provided with the package. To train an additional model, run the following command:

python -m panpe.train <config_name>

where <config_name> is a name of a configuration file, such as panpe-2layers-xrr. Configuration files are stored in the configs directory.