Standardization / Normalization for logistic regression
I was wondering if for logistic regression do we need to standardize AND normalize or simply normalize the data? What the difference? And does it apply also to regularized logistic regression as well?
Thank you in advance and sorry if someone already asked the question
Simple standardization should be sufficient, and it applies to both regularized and normal logistic regression. The reasoning is the same as for linear regression, i.e. standardization often helps to reduce the condition number of the problem and thus speed up the convergence of gradient descent.
But I would recommend first to somehow handle the missing values (equal to -999 in the dataset for project 1) and only then apply standardization.
I think standardization alone should be sufficient. Although some students reported on the forum that they applied both standardization + normalization, but I'm not sure about the reasons why normalization on top should help when all the features are already approximately on the same scale (in terms of the standard deviation) after standardization.
Standardization / Normalization for logistic regression
I was wondering if for logistic regression do we need to standardize AND normalize or simply normalize the data? What the difference? And does it apply also to regularized logistic regression as well?
Thank you in advance and sorry if someone already asked the question
Simple standardization should be sufficient, and it applies to both regularized and normal logistic regression. The reasoning is the same as for linear regression, i.e. standardization often helps to reduce the condition number of the problem and thus speed up the convergence of gradient descent.
But I would recommend first to somehow handle the missing values (equal to -999 in the dataset for project 1) and only then apply standardization.
So there is no need to normalize the data? Because that's what I'm doing after handling the missing values for logistic regression.
I think standardization alone should be sufficient. Although some students reported on the forum that they applied both standardization + normalization, but I'm not sure about the reasons why normalization on top should help when all the features are already approximately on the same scale (in terms of the standard deviation) after standardization.
Add comment