Can you please provide an stop criterion which works based on a relative error?
The issue is, that so far all provided stop conditions are scaling either with problem size or or problem content. Therefore the lib is inly handling known data efficiently, but unknown data may cause problems.
I suggest:
1 - error_n / error_(n-1)
Can you please provide an stop criterion which works based on a relative error?
The issue is, that so far all provided stop conditions are scaling either with problem size or or problem content. Therefore the lib is inly handling known data efficiently, but unknown data may cause problems.
I suggest:
1 - error_n / error_(n-1)