Open
Description
In addition to using standard metrics like rmse/mae for optimization/learning, it is also possible to simultaneously use momentum & force losses. These are especially helpful not just in reducing the error, but also to reduce lag. Work especially well with momentum & force features.
Can also help with data noise as a result of differencing performed on force & momentum features.
This would need to be implemented to work with standard ML libraries like Sklearn, Keras & Pytorch.