Feature Engineering
Models consume numeric matrices, so raw data must be transformed: categorical values encoded, skewed values transformed, dates decomposed, interactions constructed, and dimensionality reduced. The craft is encoding domain knowledge into representations a model can exploit — usually worth more accuracy than swapping model families.
This Concept is waiting for its first lesson!
Models consume numeric matrices, so raw data must be transformed: categorical values encoded, skewed values transformed, dates decomposed, interactions constructed, and dimensionality reduced. The craft is encoding domain knowledge into representations a model can exploit — usually worth more accuracy than swapping model families.
Are you a teacher? Sign in to start contributing.
Sign In