Module autodiff

Expand description

The autodiff module.

Modules§

accumulate: Accumulate trait. Reverse accumulation trait. This trait is used to reverse accumulate the gradient for different types.
gradient: Implements the gradient computation. This module contains the Gradient trait. Each implementation of wrt returns the chosen partial derivatives.
graph: The Graph (aka. tape or Wengert List). This module contains the implementation of the computation Graph. The graph is also known as a Wengert List.
graphviz: Visualisation of the Graph. This module is for visualising a Graph.
overload: Operator/function overloading. This module contains the overloaded operators and primitive functions. In Griewank and Walther - Evaluating Derivatives, they refer to this as the “elemental library”. Operations such as + and * are redefined, along with primitive functions such as sin, exp, and log. Each overload has an associated test to ensure functionality.
variable: Variables for autodiff. This module contains the implementation of the Variable structure.
vertex: Implements Vertex (nodes) for the Graph.

Graph: Struct to contain the graph (Wengert list), as a vector of Vertexs.
Variable: Struct to contain the initial variables.
Vertex: Struct defining the vertex of the computational graph.

Arity: Enumeration for the operation type. This is used to determine the number of parents. For example: - A binary operation has two parents. - A unary operation has one parent. - A nullary operation has no parents.

Accumulate: Trait to reverse accumulate the gradient for different types.
Gradient: Return the derivative/s with-respect-to the chosen variables. This allows you to get the gradient of a function with respect to any selection of variables, i.e. - a single variable, - a subset of the variables, - or all of the variables.
Log: Overload the Log trait.
Max: Overload the Max trait.
Min: Overload the Min trait.
Powf: Overload the Powf trait.
Powi: Overload the Powi trait.