Feature description
Update AdamW's sigma0 implementation for parameter specific values.
Motivation
As sigma0 represents the learning rate of AdamW, having parameter specific sigma0 values will allow for parameter specific learning rates. One such area where this will help is in the GaussianLogLikelihood function, where the sigma value of the output variable is estimated, and is conventionally small (1e-3) for conventional echem output variables ("Voltage [V]").
Possible implementation
No response
Additional context
No response
Feature description
Update AdamW's sigma0 implementation for parameter specific values.
Motivation
As
sigma0represents the learning rate of AdamW, having parameter specificsigma0values will allow for parameter specific learning rates. One such area where this will help is in theGaussianLogLikelihoodfunction, where the sigma value of the output variable is estimated, and is conventionally small (1e-3) for conventional echem output variables ("Voltage [V]").Possible implementation
No response
Additional context
No response