Notebook with experimental newton implementation #944
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
| @@ -0,0 +1,181 @@ | |||
| { | |||
There was a problem hiding this comment.
Line #7. grad = pt.linalg.solve(jac, f_x, assume_a="sym")
You're assuming jacobian is symmetrical, but that shouldn't be true in general right?
Reply via ReviewNB
There was a problem hiding this comment.
Oh, you are right. It is in the case of minimization, but might not be for different root finding problems. We should add an option for that.
There was a problem hiding this comment.
Ah, good catch! The jacobian is indeed generally asymmetrical for n>1.
At first I thought you were talking about the Hessian. That is symmetrical but with a really weird caveat for the case when things are twice differentiable but the second derivatives aren't continuous. (In these sorts of cases you can get some really weird stuff like a fractal trail that's flat at every point but ascends.)
Some intermediate code for inla