Creating an HGF and HGF agent

In this section we will cover the types of nodes, their parameters and the rules for structuring your own HGF.

Overview

states and parameters

Building principles

The following rules apply for connecting nodes, when customizing your own HGF structure:

Parameters

  • no parameters in the categorical state node

The states of Categorical input nodes and parameters

  • input value

Parameters

  • no parameters in the categorical state node

Continuous Nodes

The states of Continuous state nodes and parameters

States

  • posterior mean
  • posterior precision
  • value prediction error
  • volatility prediction error
  • prediction mean
  • prediciton volatility
  • prediction precision
  • auxiliary prediction precision

Parameters

  • evolution rate (default is 0)
  • value coupling
  • volatility coupling
  • initial mean (default is 0)
  • initital precision (default is 0)

The states of Continuous input nodes and parameters

  • input value
  • value prediction error
  • volatility prediction error
  • prediction volatility
  • prediction precision

Parameters

  • input noise (default is 0)
  • value coupling
  • volatility coupling

Binary state node rules:

  • Can only have exactly one value parent
  • Can only have excatly one value child
  • Can only have a contionus state node as value parent

states and parameters

continuous state node rules:

  • Can’t have binary input node as child
  • Can’t have binary input node as volatility child
  • Contionus state node having a binary input node as volatility child
  • Can’t have contionus input node as value child while also having volatility children
  • Can’t have the same value parent as volatility parent
  • Can’t have the same value child as volatility child

states and parameters

Categorical state node rules:

  • Can only have exactly one value child
  • Can only have categorical input node as child
  • Can only have binary state node as parents

states and parameters


This page was generated using Literate.jl.