In this slide, it is explained how we could choose whether or not to rescale a feature. My question is, suppose our dataset has 10 features, and we decide to rescale 2 of them, would it not affect the other features somehow? I mean, is it advisable to apply rescaling to all features just to be safe, or it’s safe to proceed with only the necessary ones?
Also, how do we identify when to apply mean normalization or Z-score normalization? If I apply mean norm on one feature and Z-score norm on another, what would be the possible drawbacks (if any) to my results, considering that my dataset has features with both positive and negative numbers?
Apologies for such a long question
Hello @prateekm_2709,
The rule of thumb is to scale all of them. The lower right plot is what we want to achieve, and to achieve that, one condition is that all features should be in very similar range, and that’s why the rule of thumb. The slide demonstrated and explained when we need scaling, we definitely don’t want the bottom 3 ranges, but even for the upper 2 ranges, it’s safe and better to just normalize - again, we want all features to be in very similar range.
For the scope of our specialization, it makes no difference using either of the two.
1 Like
@rmwkwok thanks for the help, it’s all clear now.
Just another thing, as I had mentioned in the question, does it make sense to use both types of normalization in a single dataset, or it’s advisable to stick to only one?
Hi @prateekm_2709 , we always stick with one normalization.
@rmwkwok got it, thanks again 
