Welcome to our Support Center

regularizer

regularizer : enum, adds a penalty to the weights to limit their growth and improve the model’s generalization.

 

Default

In default mode, you can choose between no regularization (default setting) or apply a predefined scheme: L1, L2, or L1L2.
In this configuration, the regularization coefficients l1 and l2 are fixed to 0.01 and cannot be modified.
This mode offers a simple way to introduce regularization without manually adjusting parameters.

L1

L1 regularization adds a penalty proportional to the absolute values of the weights:

This promotes sparsity by encouraging weights to become exactly zero.
When selected explicitly, the l1 coefficient can be set freely.

L2

L2 regularization adds a penalty proportional to the squared values of the weights:

This helps prevent overfitting by discouraging large weights and smoothing the model.
The l2 coefficient is user-configurable when this mode is selected.

L1L2

L1L2 combines both L1 and L2 penalties:

It balances sparsity (from L1) and weight decay (from L2), offering finer control over the regularization behavior.
Both l1 and l2 coefficients are available for customization in this mode.

Table of Contents