Abstract

The goal of machine learning is to use data to create a predictive model. Machine learning models encapsulate a function to calculate an output based on one or more inputs. It creates a model from data.

Consists of two phases:

  1. Training — the process of defining the function.
  2. Inferencing — using the model to predict new values.

Training Machine Learning Models

Consider an ML model that predicts ice cream sales based on the weather:

  • Training data comes from past observations.
    • Features ($x$) — attributes of the thing being observed (temperature; rainfall; wind speed).
      • Usually a vector: [$x_1, x_2, x_3, …$]
    • Labels ($y$) — values (the number of ice creams sold on a day).
  • Algorithms are applied to the training data to find a relationship between the features and the labels and then generalize that relationship as a calculation.
    • This is fitting the data to a function.
  • The algorithm’s output ($\hat{y}$) is a model that encapsulates the calculation it performs as a function.

Types of Machine Learning