---

name: gas-models description: "Use PyFlux GAS, GASX, GASReg, GASRank, and GAS local-level/trend models for score-driven forecasting and paired comparisons." disable-model-invocation: true metadata: disco-role: operating license: BSD 3-Clause

PyFlux GAS models

Use this sub-skill when the task is a PyFlux score-driven model: univariate GAS, exogenous GASX, dynamic GASReg, score-driven local level/trend (GASLLEV, GASLLT), or dynamic paired-comparison GASRank.

Route map

Use this sub-skill for:

• pf.GAS(data, ar, sc, family, integ=0, target=None) for univariate score-driven forecasting.
• pf.GASX(data, formula, ar, sc, family, integ=0) for GAS with exogenous regressors.
• pf.GASReg(formula, data, family) for score-driven dynamic regression coefficients.
• pf.GASLLEV(data, family, integ=0, target=None) and pf.GASLLT(data, family, integ=0, target=None) for score-driven local-level and local-linear-trend models.
• pf.GASRank(data, team_1, team_2, family, score_diff, gradient_only=False) for local paired-comparison/ranking data.

Route elsewhere:

• Kalman-style state-space wrappers that are not prefixed GAS go to ../state-space-models/.
• GARCH/EGARCH conditional volatility goes to ../volatility-models/.
• ARIMA/ARIMAX-style mean models go to ../univariate-models/.
• Do not make live HTTP NFL CSV reads a runtime dependency; use the offline GASRank fixture pattern in the workflows reference.

Default operating procedure

1. Classify the data: univariate time series, formula model with exogenous columns, local level/trend, or paired comparison.
2. Validate family/data compatibility: signed continuous series can use continuous families; count/intensity series must be nonnegative before using Poisson or Exponential-style families.
3. Fit first with a small deterministic/offline workflow and MLE unless a Bayesian posterior check is specifically needed.
4. Validate the fit by checking latent-variable values and forecast/prediction outputs for finite values, expected horizon length, and ordered interval columns when intervals are requested.
5. For GASRank, ensure every team and second-component name that will be predicted appears in the local training DataFrame.

References and smoke check

• API reference
• Offline workflows
• Troubleshooting
• Root families and inference for priors, fit methods, result objects, and posterior diagnostics.
• Root troubleshooting for install/build and legacy live-data failures shared across routes.
• Smoke helper: ../../scripts/smoke_pyflux_models.py --section gas