Some data science in Haskell

Here is a UK property price model in Haskell:

It interpolates the price of a UK house across all time from just public price paid data. It features a fairly large sparse non-linear regression problem solved with a custom implementation of the Adam optimiser (adamax variant), the ad autodiff package and a bit of wrangling in about 55 lines of code.

Note: it uses deepseq at the moment which I intend to replace with strict-containers: I had a version conflict issue due to primitives, and I don’t understand cabal well enough to solve it yet, but I’ll change it when I do.

Thank you those that helped me muddle through in this thread.


Jointly learning transformations and fitting a regression

A 5 parameter non-linear pipeline composed of a two Yeo-Johnson transforms, a sigmoid function which bisects the model, and a linear regression. It is jointly fitted using “threshold accepting” annealing which is also implemented herein: 53 lines of code.

1 Like