search

chengsoonong / mclass-sky

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

Plot learning curve of SGD and GP #165

Closed nbgl closed 7 years ago

nbgl commented 7 years ago

R^2 vs training point #.

nbgl commented 7 years ago

Progress made in ba42d23 (#169).

nbgl commented 7 years ago

Do passive learning curves second.

nbgl commented 7 years ago

I have obtained the following curve for passive learning. I am working on unifying SGD and GP and producing a better one.

From this, we can see that GP works very well on small training sets. Perhaps there is no need for SGD regression?

curve

nbgl commented 7 years ago

At long last, I produced correct curve! I take back what I said about not needing SGD regression: you can clearly see that the accuracy of SGD regression keeps increasing beyond 1.5K training points, which is where we stop being able to train GP regression.

curve2

nbgl commented 7 years ago

Completed for passive learning in 529ff9b.