What a lunkhead I am!
Dotard, a-dozing at the very nonce,
After a life spent training for the sight!
Robert Browning, Childe Roland to the Dark Tower Came
I can totally do some of version 3 in an easier way than I’d planned.
- I don’t see why each part (shots, pens, etc.) can’t be fitted separately.
- Up till last night I had the sense I’d have to estimate them jointly.
- Otherwise, the estimated relationships of GP to ability could get biased.
- Thing is, this bias can’t be that big and it’s in a third-order model feature.
- So let’s just fit components one at a time – far fewer parameters at once!
- That will mean easier debugging and diagnosis, faster convergence, etc.
Version 3’s been slow to arrive for many reasons, with which this’ll help!
- I haven’t got a lot of spare time or attention for it in the first place.
- But, also, I’ve planned around it being tougher than needed.
- I well might’ve had some results at RITSAC but for trying to fit jointly.
- I still have no schedule for getting it all done. Other concerns first.
- Simplifying nonetheless makes it look much more achievable at all.