Log transformations in Ridge and lasso

#1
Do we need to do a log transformation if we are running a ridge or lasso? I was told if I was doing regularisation I wouldn't need to. All the predictors as well as the response variable would undergo transformation.
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Well in these models the inputs are usually all converted to standard normal. I felt by the user. I have always had concerns that how do categorical variables get converted (which you can do this, but if they are >2 groups, seems very weird and confusing, right).

I have historically only used them in the logistic settings and I have moved away from them, preferring Bayesian models where the regularization comes from the user provided priors.
 

Dason

Ambassador to the humans
#3
Converting to a standardized version of a variable really makes no difference at all and typically the algorithm will take care of that part for you. But a log transform is a different beast and would need to be done manually.

Maybe OP is confusing ridge/lasso with tree/forest based methods? Monotonic transformations really don't make a difference there (at least for main effects) in terms of prediction but they still very much have an impact on any linear model based algorithm which ridge and lasso are a part of.
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
I suppose another question is why transform data - e.g., skewness, interpretatibility (scale), etc. And if it is necessary.