Generalized Stable Weights via Neural Gibbs Density

We present a generalized balancing method -- stable weights via Neural Gibbs Density -- fully available for estimating causal effects for an arbitrary mixture of discrete and continuous interventions. Our weights are trainable through back-propagation and can be obtained with neural network algorithms. In addition, we also provide a method to measure the performance of our weights by estimating the mutual information for the balanced distribution. Our method is easy to implement with any present deep learning libraries, and the weights from it can be used in most state-of-art supervised algorithms.