dataPreparation (version 1.1.1)

fast_scale: scale

Description

Perform efficient scaling on a data set.

Usage

fast_scale(data_set, scales = NULL, way = "scale", verbose = TRUE)

Value

data_set with columns scaled (or unscaled) by reference. Scaled means that each column mean will be 0 and each column standard deviation will be 1.

Arguments

data_set

Matrix, data.frame or data.table

scales

Result of function build_scales, (list, default to NULL).
To perform the same scaling on train and test, it is recommended to compute build_scales before. If it is kept to NULL, build_scales will be called.

way

should scaling or unscaling be performed? (character either "scale" or "unscale", default to "scale")

verbose

Should the algorithm talk? (Logical, default to TRUE)

Details

Scaling numeric values is useful for some machine learning algorithm such as logistic regression or neural networks.
Unscaling numeric values can be very useful for most post-model analysis to do so set way to "unscale".
This implementation of scale will be faster that scale for large data sets.

Examples

# Load data
data(adult)

# compute scales
scales <- build_scales(adult, cols = "auto", verbose = TRUE)

# Scale data set
adult <- fast_scale(adult, scales = scales, verbose = TRUE)

# Control
print(mean(adult$age)) # Almost 0
print(sd(adult$age)) # 1

# To unscale it:
adult <- fast_scale(adult, scales = scales, way = "unscale", verbose = TRUE)

# Control
print(mean(adult$age)) # About 38.6
print(sd(adult$age)) # About 13.6