About the evaluation in the neuro-symbolic paper about pybpl HOT 2 CLOSED

Hi,

Thank you for your interest in our paper. We will be releasing code for the neuro-symbolic generative model sometime in the next few months after finishing a follow-up project. The reason that our log-likelihoods look different from the DeepMind paper is that they are reporting Bernoulli pixel likelihoods on the image directly, whereas our model is reporting probabilities for the latent stroke parameters. In our upcoming paper we will be reporting analogous image log-likelihoods (marginalized over latent stroke parameters).

Thanks again for the interest.

from pybpl.

Hi @ZENGXH - we've now released the source code for our neuro-symbolic generative model: https://github.com/rfeinman/GNS-Modeling

from pybpl.