How to Scale Numerical Data

Question 1

WHICH TYPE OF ALGORITHMS BENEFIT FROM SCALING NUMERICAL VALUES TO A STANDARD RAGE? P230

Answer 1

Algorithms that use a weighted sum of the input (e.g. Linear Regression, ANN)
Algorithms that use distance measures (e.g. K-nearest neighbors, SVM)

Question 2

WHAT ARE THE TWO MOST POPULAR TECHNIQUES FOR SCALING NUMERICAL DATA? P230

Answer 2

Normalization

Standardization

Question 3

WHAT DOES NORMALIZATION DO? P230

Answer 3

Scales each input variable separately to the range 0-1

Question 4

WHY DO WE CHOOSE THE RANGE 0-1 FOR NORMALIZING? P230

Answer 4

Floating-point values have the most precision in this range

Question 5

WHAT DOES STANDARDIZATION DO? P230

Answer 5

Scales each input variable separately by subtracting the mean and dividing by STD so that the distribution has a mean of 0 and STD of 1

Question 6

FOR WHICH MODEL IS IT A CRITICAL STEP TO SCALE THE TARGET? P231

Answer 6

For regression predictive modeling problems, to make it easier to learn; most notably in the case of neural network models

Question 7

WHY DO WE NEED TO KNOW THE RANGE OF VALUES, BEFORE NORMALIZING? P232

Answer 7

Because we need all the values to be between 0 and 1, we divide all values by the maximum amount, or we use the range of change (maximum value-minimum value).
Normalization formula: y = x − min /max − min

Question 8

WHAT SHOULD WE DO IF THERE’S AN OBSERVATION HIGHER OR LOWER THAN THE MAX – MIN VALUES? P232

Answer 8

Either remove them or limit them to the pre-defined maximum and minimum values

Question 9

USING WHICH CLASS CAN WE NORMALIZE DATA IN SCIKIT-LEARN? P232

Answer 9

MinMaxScaler

Question 10

USING WHICH METHOD OF MINMAXSCALER CAN WE REVERSE THE TRANSFORMATION? P233

Answer 10

Inverse_transform ()

Question 11

WHAT IS THE ASSUMPTION OF STANDARDIZATION? P234

Answer 11

That the observations fit a Gaussian distribution, with a well-behaved mean and STD.

Question 12

WHAT’S ANOTHER NAME FOR STANDARDIZATION? P234

Answer 12

Center scaling

Question 13

WHAT IS SCIKIT-LEARN’S CLASS FOR STANDARDIZATION? P234

Answer 13

StandardScaler

Question 14

WHEN IS IT BETTER TO NORMALIZE AND WHEN IS IT BETTER TO STANDARDIZE? P244

Answer 14

Whether input variables require scaling depends on the specifics of your problem and of each variable. You may have a sequence of quantities as inputs, such as prices or temperatures. If the distribution of the quantity is normal, then it should be standardized, otherwise, the data should be normalized. This applies if the range of quantity values is large (10s, 100s, etc.) or small (0.01, 0.0001).