Just skimming, it looks as though they just want a way to have standard deviation depend linearly on a regressor (i.e. modelling heteroskedasticity)? I don't know if their method works, but if you just want to do that, it's much easier in a Bayesian setting.
Let's talk CPEM, or Continuous Parameter Estimation Model. Supposedly a clever trick to get the interaction term from linear regression using dot product and standardized variables. But does it really? Stats math people, what say you? Prior = low. https://www.gwern.net/docs/statistics/2005-gorsuch.pdf …pic.twitter.com/hSCxPzxFOa
-
-
-
Here's a simple example with simulated data, using pymc3.pic.twitter.com/HNmxuSVVlI
-
@jonatanpallesen did this simulation. It seems to show that CPEM works for some simulated data and not others. It depends on some data feature, correlation or some other structure. Not sure. http://jsmp.dk/files/cpem_sim
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.