This pull request adds regularization to the epitope activities. The regularization is a Pseudo-Huber function for positive activities, and zero for negative activities. This helps keep activities <= 0 for epitopes that are in unimportant.
Empirically, this improves fitting speed a bit and seems to otherwise give comparable-ish results. The default weight is 1.
This pull request adds regularization to the epitope activities. The regularization is a Pseudo-Huber function for positive activities, and zero for negative activities. This helps keep activities <= 0 for epitopes that are in unimportant.
Empirically, this improves fitting speed a bit and seems to otherwise give comparable-ish results. The default weight is 1.