Exploratory Data Analysis

Understanding a dataset and turning its contents into numeric features a model can learn from. These pages are vendor-neutral.

Categorical data encoding covers ordinal, one-hot and binary encoding, and when each preserves or destroys the meaning of the original categories.

Three pages deal with extracting features from unstructured data: text — bag of words, n-grams, TF-IDF, embeddings, stemming — and images and speech — pixel values, edge detection, MFCC and the deep learning alternatives.

Dimensionality reduction and feature selection covers what to do when there are too many features: filter, wrapper and embedded selection methods, and extraction methods such as PCA, LDA and t-SNE.

Cleaning, imputation and scaling are covered under data engineering.