User guide

The user guide explains how to do spectral decomposition with ns-sdn, one question at a time. Each page gives the relevant part of the model in its standard notation, runs a verified example, and ends with the nuances that are easy to get wrong. The API reference has the exhaustive argument lists; the guide is where to learn what each path means and when to trust it.

New to the library? Read Quickstart first, then The model and The decomposition. Those two pages explain what the network computes and what the result hands back, which every other page assumes.

Find the right page

Question	Page
What does the network compute, head by head, and how is it trained?	The model
What is each path on the result, and how do they add up?	The decomposition
One-step versus recursive, filtered versus predicted: which number means what?	Forecasting
How do I summarize sixteen components, and test a regime-change claim?	The spectral state
What shape do inputs take? How are returned arrays indexed and scaled?	Conventions
What behaves differently from what the papers lead me to expect?	Nuances

The model

• The model
• The decomposition

Using a fit

• Forecasting
• The spectral state

Foundations

• Conventions
• Nuances