Data Normalization
Scaling schemes
Each scheme brings values of different features into comparable ranges, but each of them preserve different types of information (and distort others) as they do so
Standard Scaler
Scale the features such that each feature’s Gaussian distribution is centered around 0 and it’s standard deviation is 1.
Normalizer
Looks for all the feature and divide them by the magnitude.
MinMaxScaler
Shinkrs or stretches data in a range of value. Some example of range values are: [0,1], [-1,1], for pixels is [0,255]
(value - feature_min)/(feature_max - feature_min)
This scheme works much better in certain cases where StandardScaler might not work well. For example, if the standard deviations are very small for features, StandardScaler is highly sensitive to tiny changes between standard deviations of different features, but MinMaxScaler is very robust. Also, for features with highly skewed distributions, or sparse cases where each feature has a lot of zeros that moves the distribution away from a Gaussian, MinMaxScaler is a better choice.