Regularizers

The Regularizer implementations add a parameter penalty to the training objective. They are configured under trainer.regularizer and applied while the Trainer trainer computes the total loss.

Available implementations

none or no (NoReg) disable regularization.
l1 (L1Reg) applies an L1 penalty.
l2 (L2Reg) applies an L2 penalty.

Example

trainer:
  regularizer:
    regularizer_type: l2
    weight: 0.0001

API reference

See also xanesnet.regularizers and xanesnet/schemas/regularizers/ for complete fields and defaults.