API

dtfit API reference

Complete reference for the public dtfit API -- every name exported from the top-level package and the dtfit.diagnostics submodule. Internals (anything under a _-prefixed module) are not part of the public contract.

For the ideas behind these functions read ../guides/; for the math read ../methods/. This reference is for looking up exact signatures, arguments, return types, and behavior.

Looking for the experimental adaptations (the Fourier, Chebyshev and Laguerre bases, fit_aggregated, fit_aligned, the weak-form fitters, LocalTimeFilter)? Those live in the separate dtfit-experimental package -- see ../experimental/adaptations-api.md. The InformationFilter fusion primitive lives there too (from dtfit_experimental import InformationFilter), not in the stable streaming surface. This page covers the stable dtfit API.

Conventions

• Fitters take an Original or an Image directly, or plain 1-D NumPy arrays x, y (from which the batch fitters build an Original), and a model as a sympy-style expression string, a sympy.Expr, or a callable f(x, *params), plus the name of its main variable, e.g. fit("a*exp(b*x)", Original(x, y), "x", basis="legendre").
• Parameters are the free symbols of the expression (everything except the variable), and results are ordered by sorted parameter name -- a stable layout used everywhere. So "a*exp(b*x)" has parameters [a, b] in that order.
• Batch fitters return a FittingResult; streaming filters expose partial_fit / predict / params_.
• Keyword-only arguments appear after * in the signatures below.

The public surface, by area

Area	Names	Page
Batch fitting	fit, Original, Image, order_for, fit(basis="legendre"), fit(basis="block") (coverage, fft_frequency_seed and the Image analytics noise_sigma / effective_order / decay / test_equal / test_structure in dtfit.image)	fitting.md
Result type	FittingResult	types.md
sklearn estimator	NonlineRegressor (in dtfit.sklearn)	estimator.md
Forecasting	auto_forecast, ForecastResult	auto.md
Model framework	models, suggest_models (Model, register, unregister and the catalog families in dtfit.models)	models.md
Reference method	fit_dsb, find_degree (in dtfit_legacy.dsb)	dsb.md
Stochastic series	stochastic (fit_stochastic, StochasticModel, StochasticFilter, SecondOrderImage, SecondOrderStream and the estimators in dtfit.stochastic; Stochastic in dtfit.models)	stochastic.md
Streaming / online	ImageFilter, ImageFilter(basis="legendre"), ImageFilter(basis="block") (DriftDetector in dtfit.streaming)	streaming.md
Streams and scale	ImageStream, fit_many (FittingProblem, assemble, legendre_transfer, block_transfer in dtfit.image)	scaling.md
Diagnostics	diagnostics (fit_report, residual_diagnostics, residual_stats, FitDisplay, ResidualsDisplay)	diagnostics.md
Logging	enable_logging, logger (in dtfit.log)	below

The top level holds fifteen names and three subpackages: Original, Image, ImageStream, ImageFilter, fit, fit(basis="legendre"), fit(basis="block"), ImageFilter(basis="legendre"), ImageFilter(basis="block"), order_for, fit_many, suggest_models, auto_forecast, FittingResult, ForecastResult, plus models, stochastic and diagnostics. Everything else is reached through its own module, as the import map below shows.

Import map

# batch fitting
from dtfit import fit, Original, Image, order_for, FittingResult
from dtfit.image import coverage, fft_frequency_seed

# forecasting
from dtfit import auto_forecast, ForecastResult

# model framework
from dtfit import models, suggest_models
from dtfit.models import Model, register, unregister, resolve_model

# the reference method (the exact-balance ancestor, not part of fit)
from dtfit_legacy.dsb import fit_dsb, find_degree

# stochastic series (characterize / forecast / generate / track random data)
from dtfit import stochastic
from dtfit.models import Stochastic
from dtfit.stochastic import (
    fit_stochastic, StochasticModel, StochasticFilter, SecondOrderImage,
    SecondOrderStream, hurst_spectral, ar1_reversion, garch_persistence,
    cycle_period, decompose_trend_cycle, dickey_fuller, ar_order, fit_ar,
    fractional_difference, FORECASTERS)

# sklearn estimator (the only part of dtfit that imports scikit-learn)
from dtfit.sklearn import NonlineRegressor

# streaming
from dtfit import ImageFilter
from dtfit.streaming import DriftDetector

# streams and scale
from dtfit import ImageStream, fit_many
from dtfit.image import (FittingProblem, assemble, legendre_transfer,
                         block_transfer)

# Image analytics (also reachable as Image methods)
from dtfit.image import (noise_sigma, effective_order, decay,
                         test_equal, test_structure)

# diagnostics (submodule, not top-level - sklearn convention)
from dtfit.diagnostics import (fit_report, residual_diagnostics,
                               residual_stats, FitDisplay, ResidualsDisplay)

# logging
from dtfit.log import enable_logging, logger

Logging

dtfit is silent by default. Opt in to see what the solvers are doing:

from dtfit.log import enable_logging
enable_logging()              # INFO-level chatter from the methods
enable_logging(level="DEBUG") # more detail

• dtfit.log.enable_logging(level="INFO") -- attach a handler to the library logger and set its level. Call once at startup.
• logger -- the underlying logging.Logger ("dtfit"), if you want to wire it into your own logging configuration instead.

Internally the methods emit progress through a small echo helper (window counts, selected polynomial degree, fitted coefficients); none of it prints unless you enable logging.

A 30-second example

import numpy as np
from dtfit import fit, Original
from dtfit.diagnostics import fit_report

x = np.linspace(0, 4, 200)
y = 0.5 + 2.0 * np.exp(0.5 * x) + np.random.default_rng(0).normal(0, 0.2, x.size)

res = fit("a0 + a1*exp(a2*x)", Original(x, y), "x", basis="legendre", order=6)
print(res.summary())                 # parameters +/- standard errors
print(fit_report(res, x, y)["r2"])   # goodness of fit
y_hat = res.predict(x)               # evaluate the fitted model