What makes Z-score std a little bit better in unit 3.7 Feature Normalization (Parts 1-2). #47
-
|
From my understanding, It might be ... there are more chance to get gradient from both sides (+), (-) because of zero centering? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
Yes that's right. You can have positive and negative values with the gradient and the weights. So there are 4 different settings negative weight & negative input, negative weight & positive input, positive weight & negative input, positive weight & positive input. If you normalize to [0, 1] range, you can only have negative weight & positive input, and positive weight & positive input. |
Beta Was this translation helpful? Give feedback.
-
|
wow short and clean thanks :) |
Beta Was this translation helpful? Give feedback.
Yes that's right. You can have positive and negative values with the gradient and the weights. So there are 4 different settings negative weight & negative input, negative weight & positive input, positive weight & negative input, positive weight & positive input. If you normalize to [0, 1] range, you can only have negative weight & positive input, and positive weight & positive input.