Featurization concepts in AML
In order to provide the best model, regardless of whether AutoML is being leveraged, an important step in model creation is the engineering features. AutoML in AMLS will default to leverage featurization. This can be disabled in the UI as well as the SDK if the feature engineering step has already been accomplished. These featurization transformations on your dataset can not only be enabled or disabled but they can also be customized or excluded from specific columns. There are several featurization steps applied to your dataset based on the type of column, and the column’s data type.
During training, AutoML leverages scaling or normalization to ensure model performance. AutoML leverages a variety of techniques, including scaling to unit variance, scaling by quantile range, scaling by the maximum absolute value, scaling by a column’s minimum and maximum, by applying Principle Component Analysis (PCA) for dimensionality reduction, Singular...
