System and SystemParams

System and SystemParams#

A System is the unit of model specification: one declaration of the state factors, the observation modalities, the control nodes and the objective. SystemParams is the numeric state that goes with it: the transition, likelihood and prior of each variable, the preferences, and the parameters of anything learned. Everything else in the library takes or returns a System. A System may be discrete, continuous or mixed. The entries a SystemParams holds are described under Model components.

The declaration#

System

Static structure of a System: discrete, continuous or mixed.

SystemParams

All numeric state of a System: generative tensors, designations and the closed-loop factors.

Structure of a built System#

These are carried on a System and read back from it, more often than they are written.

LearnState

Optimizer state for an agent's gradient-learnable parts: learn's explicit carry.

Ports

A System's ports and their types (System.ports()): each maps a port's name to (LEVEL, levels) or (FLOAT, dimension).

Subsystem

One component's identity inside a merged composite, carried on System.subsystems.

PartParams

One leaf's own planner parameters inside a composite, SystemParams.parts[leaf].