Declarative DeepProblog is a declarative extension to DeepProblog and all neuro-symbolic languages that rely on neural predicates.
This problem relies on deepproblog-dev (available from https://github.com/ML-KULeuven/deepproblog-dev), which provides a complete setup guide.
An implementation is also provided in the separate deepproblog.zip. Please follow the installation instructions.
This also includes the added predicates needed.
Please install the requirements.txt afterwards.
To run the declarative extension, navigate to any of the examples and run distr_generative.py.
Run python distr_generative.py --h to see all available options.
To run our experiments on MNIST, see the README in the MNIST dir.
Two things are necessary:
- Encoder and decoder networks that map your entities into latent space (and back), and
- The DPL model formulation.