A variational autoencoder to learn the DAG of a structural equations model
DAG-GNN implements a variational autoencoder to learn the DAG of a structural equations model in Tensorflow (in R). The the method of the same name has been published by Yu et al. recently.
The basic idea of the method is as simple as beautiful. Cast the structure learning problem into a continuous optimization problem using a variational autoencoder and enforce acyclicity with an equality constraint.
You can install the packagefrom GitHub
remotes::install_github("dirmeier/dag-gnn")Simon Dirmeier simon.dirmeier @ web.de