mnist example get loss nan

[aichifandefan]

when i run the mnist example i get the loss is nan, is the ToPoincare cause result?

[aichifandefan]

when i commented out the code for self.tp and self.mlr and turn in Euclidean space to classificiation it not get nan result

[07Agarg]

Hi @leymir , I am also getting nan in mnist example. How to resolve this?

[aichifandefan]

@07Agarg When I adjust the c to 0.05 acorroding to the paper no longer get nan.

[aichifandefan]

@KhrulkovV @leymir I still have a confusion, in the paper the c means the radius of the ball, but in the source code it means ball negative curvature , is this a wrong comment?

[07Agarg]

@07Agarg When I adjust the c to 0.05 acorroding to the paper no longer get nan.

Solved my issue. Thanks @aichifandefan !