Terminology
Machine Learning (ML): a subset of Artificial Intelligence (AI). ML provides algorithms with the ability to improve without being explicitly programmed. It takes in a dataset of inputs and outputs and adjusts its parameters to accurately predict an output given a new input.
Black-Box Algorithm: ML/AI algorithms in which the actual program is not examined. There is no insight to how the model converts inputs to outputs.
Loss function: a function which evaluates how well a ML algorithm models a dataset (low value: good, high value: bad). It maps an event associated with the model's variables to determine the "cost" of the event.
Optimizing: the process of minimizing a loss function.
Gradient descent: an optimization algorithm which finds the local minimum of a function (in our example, the loss function). To do so, an initial value is chosen and steps are taken down the negative of the gradient/slope calculated at the current point.
Continuous features: numeric variables with an infinite number of possible inputs (e.g. age, time)
Categorical features: variables with a limited number of possible inputs (e.g. gender, country, race )
One-Hot Encoding: a method which converts categorical features to binary numerical inputs. Machine Learning models only take numbers as an input, so one-hot encoding is used to convert strings (text) to numbers in a way that the model can interpret.
