Open jlevy44 opened 5 years ago
At the moment, many of our estimators build on top of any model that supports appropriate fit
and predict
methods, which could include custom models built on top of PyTorch. So, for example, you could pass a custom model class (where fit
trains a PyTorch network and predict
feeds the input forward through the trained network to get the result) as one of the arguments to the DMLCateEstimator
initializer.
However, our DeepIVEstimator
(which explicitly uses deep neural networks to perform instrumental variables estimation) is currently built on top of Keras with no corresponding way to use PyTorch. It would certainly be worthwhile to investigate whether we could include a PyTorch option as well, although it's not on our short-term roadmap. Thanks for the suggestion!
Will pytorch support be offered for this package?
I can think of a number of biomedical applications that could largely benefit from these approaches.