Hey Sebastian, thanks for the video. As for the cons for using the Variance threshold, could we say that it doesn't take into account the scale of the variables ? And thus it would only be applicable if we normalized our dataset.
Yeah, if you don't have a normalized dataset, it would be very tricky to find a good variance threshold that works well across features. For binary variables, it is somewhat easier to reason what a good threshold might be, but for continuous variables, it's extremely hard to come up with a good number as a threshold, imho.
@@SebastianRaschka I started to apply variance thresholding to a dataset and was also wondering about scaling and non-binary features. It turns out variance threshold returns 1 for all normalized data. So variance threshold simply measures the standard deviation.
Thank youuuuuuuu
does Weka have something like variance threshold remover?
Hey Sebastian, thanks for the video. As for the cons for using the Variance threshold, could we say that it doesn't take into account the scale of the variables ? And thus it would only be applicable if we normalized our dataset.
Yeah, if you don't have a normalized dataset, it would be very tricky to find a good variance threshold that works well across features. For binary variables, it is somewhat easier to reason what a good threshold might be, but for continuous variables, it's extremely hard to come up with a good number as a threshold, imho.
@@SebastianRaschka I started to apply variance thresholding to a dataset and was also wondering about scaling and non-binary features. It turns out variance threshold returns 1 for all normalized data. So variance threshold simply measures the standard deviation.