Getting started#
ns-sdn is one model behind one pattern. NSSDN
holds the series and the specification, fit() trains the network, and
the NSSDNResult it returns is a
frozen record of everything the fit is: the decomposition as arrays, the
one-step fit and its residuals, the trained network behind a numpy
interface, and a summary that states what the numbers assume. Learn the
pattern once and the whole library reads the same way; readers of
cultivars already know it.
The three pages below take about fifteen minutes in order. Readers who know the model from the papers can start at the quickstart; readers who want to know why the library is shaped the way it is can start at the concepts.
Supported Pythons, the torch dependency, optional extras, pinning, and a development checkout.
Fit a decomposition to a series with a known answer, read every path, run the one-step and recursive protocols, compute the spectral state, compare two fits.
Models, results, summaries, the two alignments, the training schedule, data conventions, and the error hierarchy.
Where next#
After these pages, the user guide derives the model and explains how to read what it emits, the examples run the published DGS10 analysis on live FRED data, and the API reference documents every public class with its mathematics and a runnable example.