Custom cost function replicating p=2 doesn't match inbuilt?

[JPTK]

Hi there,

Love the package. One thing I noticed is that I wanted to try using a few custom cost functions. One I used: C(x,y) = 0.5*|| x - y ||^2 only gets the same as the inbuilt p=2 for some sample points, but can occasionally disagree.

Any thoughts on this?

Many thanks,
Jake

[jeanfeydy]

Hi @JPTK,

Thanks for your interest in the library!
This is quite surprising: the only case that I can think of is the computation of squared distances in the "tensorized" backend that relies on the polar identity |x-y|^2 = x^2 + y^2 - 2 * (x | y) and may induce small differences due to float precision errors.

Could you maybe provide me with a small example?

Best regards,
Jean