Machine Learning link functions for a custom ParametricRegressionFitter

[bluemoo]

Hello, I have a custom ParametricRegressionFitter. I am confident in using the Weibull distribution to model the event times of my data. However, the features I have available need to be transformed (both scaling and in some cases non-linear transformations need to be applied) before they can be used to calculate the distribution parameters. Currently I do this by hand with a best guess as to the approach, but I have a lot of data, and would like to be able to use some form of automated approach to calculate the best transformation. I have been imagining training a Neural Network to perform the transformations. Is this possible? Is my formulation of the problem even possible? Is there any existing work on this approach?

Thank you.