Closed aelkholy closed 4 years ago
I guess that you mean regularization. Check out LASSO, which has L1 regularization for linear regression.
Additional work is needed for logistic/maxent. It is a big work. It is not just an option. It is whole new algorithms :)
Can we have the option for L1 normalization in the linear / logistic / maxent algorithms? This would give us a way to zero out the weights, especially when using bag of words with a sparse vector.