I took a stab at simplifying the library. It's not there yet, but we're getting closer I think. Basically I split it into two functions: deriv_features and build_features. Deriv_features calculates all derivatives (now using autodiff but we can try and collocate and then use finite diff) and build_features takes in the prediction and the derivatives to build theta, the feature matrix. Main motivation behind this is that you can use build_features for for classical methods (haven't tested yet) so if you build a custom library you can use it with classical methods as well.
Hi all,
I took a stab at simplifying the library. It's not there yet, but we're getting closer I think. Basically I split it into two functions: deriv_features and build_features. Deriv_features calculates all derivatives (now using autodiff but we can try and collocate and then use finite diff) and build_features takes in the prediction and the derivatives to build theta, the feature matrix. Main motivation behind this is that you can use build_features for for classical methods (haven't tested yet) so if you build a custom library you can use it with classical methods as well.
Let me know what you think?