Scale Data for Machine Learning

Scaling (inputs and outputs) can improve the training process for machine learning. A common scaling technique is to divide by the standard deviation and shift the mean to 0. Another common scaling approach is to adjust all of the data to a range of 0 to 1 or -1 to 1. Each data column is scaled individually.

0:00 Overview
0:43 Equations
2:35 Scale 1D
4:37 Import Data
6:31 Split Data
7:22 Sklearn Standard Scaler
8:22 Scaling Factors
9:44 Transform Test Data
11:00 Numpy Array to DataFrame
12:31 Minmax Scaler
15:20 Inverse Transform
16:07 TCLab Histogram
18:16 Scale Data
20:42 Train Neural Network
27:43 Unscaled Neural Network
31:25 Summary

There are different methods for scaling that are important based on the presence of outliers or statistical properties of the data. Two primary methods for scaling are a standard scaler (scale by the standard deviation) and a min-max (e.g. 0-1) scaler. For classifiers and regressor such as neural networks, most of the data should be between 0 and 1 or -1 and 1.

Machine Learning for Engineers: https://apmonitor.com/pds
Data Scaling: https://apmonitor.com/pds/index.php/M...