convergence values
In the current implementation of v0.5, it is not possible to fed in convergence values, I guess. This needs to be thought about.
Comments (7)
-
repo owner -
reporter - changed status to resolved
That works. Thanks!
-
repo owner - changed status to open
The documentation doesn't match the current version of the code. I think we need to move the setting of the convergence parameters back into the
LossFunction
itself; that way the user can define their own criteria if they so choose.Right now, I think it is going in through
model.fit
which the user does not usually use --- I can't see a way to put them in throughcalc.train
. -
reporter - changed status to resolved
The documentation is updated. convergence parameters also now exist in the LossFunction method.
-
repo owner Can you supply the name of the commit where this was addressed? @akhorshi
-
reporter - changed status to open
Sorry, by documentation I meant docstring :) Opened back this issue to discuss it in person.
-
repo owner - changed status to resolved
I think it's all fixed up now. Note I expanded the description of the default parameters slightly in commit 4b4fa09.
- Log in to comment
What about this?
http://amp.readthedocs.io/en/latest/useamp.html#adjusting-convergence-parameters