Okay, I'm going to move all the HMoG stuff to its own application for now, and try to clean up the graphical code base now. All the stuff for linear gaussian harmoniums seems pretty solid, but HMoG is experimental

Okay, so HMoGs are generally working, but there's lingering numerical issues that i'm still trying to sort out

Switched to stable summation algorithms in a number of cases. Not sure if it matters but I'm a little desperate at the moment

Okay things are compiling and half working... I'd really like to clean up the type signatures though

A tiny bit more cleanup. NB: UndecidableInstance loops have been popping up recently. Don't fully understand them, but keep an eye out

Working away at a better Guassian implementation that will achieve everything I want in the graphical model space...

Okay, well initial rebuild of transition functions works now. Turns out it was worthless for performance, but maybe it's still a path forward...

Okay, seems like I have a good implementation of block matrix inversion now. (Maybe need to clean up types a bit)

Working on Gaussian implementation bugs. Everything seems fine except 'Scale'. Switching scale to an average diagonal representation rather than a sum.