search

chengsoonong / mclass-sky

Multiclass methods for astronomical data
BSD 3-Clause "New" or "Revised" License
9 stars 4 forks source link

Do additional training points always decrease uncertainty? #198

Open nbgl opened 7 years ago

nbgl commented 7 years ago

Say I have a predictor p trained on data points x1, …, xn. This predictor has a variance of v at t.

Say I also have a predictor p' trained on x1, …, xn+1. This has a variance of v' at t.

Is it necessarily true that v' ≤ v? My conjecture is yes, but I don’t seem to be able to prove it.

nbgl commented 7 years ago

I’m still curious about this. I wonder whether there’s a general result