---

name: pyflux description: "Use PyFlux 0.4.17 for time-series modeling, forecasting, probabilistic inference, ARIMA/GARCH/GAS/state-space/VAR/GPNARX workflows, and package-specific troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: BSD 3-Clause

PyFlux Repo Skill

Use this skill when a coding or research agent needs package-specific guidance for PyFlux time-series modeling, forecasting, probabilistic inference, diagnostics, or troubleshooting.

PyFlux 0.4.17 is a legacy probabilistic time-series library with a unified model lifecycle: build a model, inspect/adjust latent-variable priors, fit with a supported inference method, then forecast, backtest, sample, or run posterior predictive checks. Prefer a Python 3.7-era scientific stack for reliable use; newer Python/Numpy/Pandas combinations may fail during install or Cython extension import.

Quick start

Install the package in a compatible environment, then run a safe import check:

pip install pyflux
python scripts/check_pyflux_env.py

Minimal model lifecycle:

import numpy as np
import pyflux as pf

y = np.cumsum(np.random.normal(size=100))
model = pf.ARIMA(data=y, ar=2, ma=1, family=pf.Normal())
print(model.latent_variables)
result = model.fit("MLE")
result.summary()
forecast = model.predict(h=5, intervals=True)

Route by task

• Use sub-skills/univariate-models/ for ARIMA, ARIMAX, and NNAR single-series forecasts, exogenous regressors, AR/MA/integration choices, ARIMAX formula handling, prediction intervals, posterior samples, and PPCs.
• Use sub-skills/volatility-models/ for GARCH, EGARCH, EGARCHM, LMEGARCH, SEGARCH, SEGARCHM, EGARCHMReg, leverage terms, in-mean volatility, and return-series volatility forecasting.
• Use sub-skills/gas-models/ for GAS, GASX, GASReg, GASLLEV, GASLLT, and GASRank score-driven forecasting, local-level/trend models, and paired comparisons.
• Use sub-skills/state-space-models/ for LLEV, LLT, NLLEV, NLLT, DAR, DynReg, NDynReg, DynamicGLM, LocalLevel, and LocalTrend state-space or dynamic-regression workflows.
• Use sub-skills/multivariate-models/ for VAR, GPNARX, Gaussian-process kernels, multivariate DataFrame layout, and VAR-vs-kernel routing.

Shared references

• references/model-catalog.md: package taxonomy, public classes, and which sub-skill owns each route.
• references/families-and-inference.md: cross-cutting families, priors, fit methods, result objects, prediction diagnostics, and Aggregate model-combination guidance.
• references/troubleshooting.md: install/build, legacy pandas/network examples, ARIMAX formula whitespace, inference, convergence, data-shape, and unsupported/internal-surface fixes.
• references/repo-provenance.md: source snapshot, evidence paths, and refresh baseline.

Shared helpers

• scripts/check_pyflux_env.py: import/package/API check with no network access.
• scripts/smoke_pyflux_models.py: synthetic CPU smoke checks. Run --section all or one of univariate, volatility, gas, state-space, or multivariate.

Example:

python scripts/smoke_pyflux_models.py --section gas

Common operating procedure

1. Confirm the environment with the import helper before debugging model code.
2. Pick the route from the task: mean model, volatility model, score-driven model, state-space model, or multivariate/kernel model.
3. Build a small local synthetic fixture first; do not depend on live Yahoo/FRED/NFL examples for validation.
4. Inspect model.latent_variables, then adjust priors only when the task needs Bayesian or regularized inference.
5. Start with the fastest default fit (MLE for most models, OLS for VAR, BBVI for NNAR and non-Gaussian state-space classes).
6. Validate with summary(), predict_is(), finite latent variables, and forecast horizon shape before using long horizons or Bayesian sampling.
7. Use sample(), ppc(), plot_sample(), and plot_ppc() only after BBVI or M-H fits.

Safety and self-containment

This generated skill is self-contained. Runtime references and helper scripts live inside this skill directory and do not require opening the original repository checkout. Original tests and docs were used as evidence only; use the bundled routes, references, and synthetic helpers for future work.