Closed zakajd closed 3 years ago
regarding parameter order in loss. in PyTorch, the convention is to pass prediction first, target (true) second. the order is the same for every loss, it may not be documented everywhere but I don't seem it as an issue
For now most classes don't have good documantetion.
For example losses don't give a hint which parameter should go first. Propose to open a new PR and fix this