Domains

Domains -- the method by application

The experiment notebooks are organised by what they establish. These pages read the same results by application: for each domain, what is asked of an estimator there, which established methods it is scored against, what the notebooks measure, and where dtfit loses. NLLS is the reference throughout; where theory gives the image no advantage, parity is reported as parity.

Domain	dtfit routes	Scored against	Notebooks
Parameter estimation	fit in the Legendre and block bases, the robust image, basis="auto"	SciPy NLLS, robust NLLS, the NIST certified values	15, 14, 31
Forecasting	a declared structure fitted from the image, auto_forecast	random walk, seasonal naive, drift, ETS, Theta, (S)ARIMA, MLP, the published LTSF numbers	32
Embedded control	ImageFilter in both bases, the per-sample innovation test, FilterBank	a parameter-space EKF over a grid of settings, RLS, Kalman-CA, a sliding-window refit	21, 24
Big data	Image.merge, ImageStream, fit_many, the channel projection on the GPU	the resident fit, a serial loop, polynomial surrogates	22
Real-time GPS	the streaming Legendre and block trackers, IMU regressors, the robust window	Kalman-CA, a coordinated-turn EKF, an IMM, a Huber-hardened Kalman	33, 34
GPS hardware rig	the on-chip float32 filter, the host trackers on recorded drives	Kalman-CA, CT-EKF on the same fixes	rig
Stochastic series	the second-order image, the regime router, the merged forecaster	OLS AR(1), GARCH quasi-MLE, R/S, DFA, ETS, ARIMA, the random walk	23, 36
Archive showcase	ImageStream, fits from stored images, the image on the wire	numpy.linalg.lstsq on the raw rows, MIDAS velocities, NOAA normals	35