I was wondering if anyone here can explain the difference between the $L1$, $L2$ and $Max$ normalization mode in sklearn.preprocessing.normalize() module?
Having read the documentation, I couldn't realize the difference!
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
1
The options lead to different normalizations. if $x$ is the vector of covariates of length $n$, and say that the normalized vector is $y = x / z$ then the three options denote what to use for $z$:
- L1: $z = \| x\|_1 = \sum_{i=1}^n |x_i|$
- L2: $z = \| x\|_2 = \sqrt{\sum_{i=1}^n x_i^2}$
- Max: $z = \|x \|_\infty = \max |x_i|$
Edit: previously, using Max does not take absolute values first, so it is not equal to the $l_\infty$ norm -- however, that seems to have been updated and now it is equal to the infinity norm
-
$\begingroup$ If Max does not take absolute values first, would you get strange results when all the $x_i$ are negative, or worse if one or more were 0 and the rest negative? $\endgroup$– HenryFeb 16, 2022 at 11:56