A small PyTorch library making modelling of Hierarchical Probabilistic Graphical Models easier.
(the buzzwords are: Restricted Boltzmann Machine, Deep Belief Network, Deep Boltzmann Machine, Helmholtz Machine).

Developed together with Hendrik Elvers in a practical course "Beyond Deep Learning - Uncertainty Aware Models" at TU Munich.

With our library one can model arbitrary deep Probabilistic Graphical Models based on distributions from the Exponential Family (, with Bernoulli, Gaussian or Categorial distributed units currently implemented.
The Interaction layers can use any torch.nn.Module internally. So beyond fully connected weight layers, for example also convolutional layers are enabled.

This is a proof-of-concept for a collaboratively filled map of Corona-Test-Sites. People can register via their email and then add new test sites, so that you have an overview on the closest (and potentially cheapest) ones.
Due to time reasons this was not pursued, but I think it is a nice tool and could still be "really" published somehow.
Currently based on OpenStreetMap tile servers, thanks for providing those for such PoCs!