0
$\begingroup$

I am using the SVM function provided by scikit-learn. I would like to know whether I need to perform standardization before fitting the model. As I know, LibSVM tends to require pre-processing the data. I am not sure whether scikit-learn automatically normalizes the data instead of expecting us to handle it ourselves.

$\endgroup$
5
$\begingroup$

scikit learn does not standardize data, but it does offer utilities for you to standardize your input data yourself: http://scikit-learn.org/stable/modules/preprocessing.html

the rule of thumb is to standardize if your data aren't related. That is, if channel X is not a function of channel Y, you should standardize

Qualitatively, think about it this way, SVM 'creates a hyperplane' to separate data into categories; if the data are skewed too far in one axis, that will make it harder to draw a plane to separate them

$\endgroup$
1
$\begingroup$

By default, for any methods that use gradient descent or feature combination (e.g. PCA), I scale my data if the orders of magnitude in the features are different from each other. The optimization is easier when the fitted parameters are not too far from zero in any direction. Scaling you data doesn't matter for tree-based methods

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.