Classification

Overview

Classification is the process of assigning one of a finite set of labels to a point (or points) in a dataset.

Types

The types of classification algorithms can roughly be divided into two categories.

  • Linear - models were the first and easiest to implement. They laid the foundation for machine learning in general. Even though linear patterns are somewhat rare in real life, the linear based algorithms can generally be made to recognize non-linear patterns using kernel methods or feature extraction.
  • Non Linear

There are multiple algorithms that can be used to learn classification problems.

Response Function

The response function the case of classification maps each input to a finite set of outputs, often labels, as in the following.

let labels = ['apple', 'orange', 'cherry'];


How the classes are encoded is often dependent on the type of algorithm. For instance, the previous set of three labels could also be encoded as column vectors.
{% apple = \begin{bmatrix} 1 \\ 0 \\ 0 \\ \end{bmatrix} %}
{% orange = \begin{bmatrix} 0 \\ 1 \\ 0 \\ \end{bmatrix} %}
{% cherry = \begin{bmatrix} 0 \\ 0 \\ 1 \\ \end{bmatrix} %}

Cross Entropy Loss

The cross entropy loss function is a common loss function used in deep learning of classification problems.