import pyfixest as pf
fit = pf.feglm("Y ~ X1", pf.get_data(), family="gaussian")
fit.working_state.mu[:3]array([-1.08165392, -0.08156808, -1.08165392])
Final GLM IRLS state in within scale.
working_weights are the IRLS weights of the last iteration, W_i = w_i / (g'(mu_i)^2 V(mu_i)), where w_i is the user-supplied observation weight (one when unweighted). They therefore already contain the observation weights, which live separately and unchanged in ObservationWeights; nothing downstream multiplies by w again. For the Gaussian family W equals w. Square-root weighted arrays are solver-local temporaries and deliberately absent.
| Name | Type | Description | Default |
|---|---|---|---|
| working_response_within | NDArray[np.float64] | Final within-scale working response, shape (n_rows,). | required |
| design_within | NDArray[np.float64] | Demeaned working regressors after removing collinear columns, shape (n_rows, n_coefficients). | required |
| working_weights | NDArray[np.float64] | Final IRLS weights including observation weights, shape (n_rows,). | required |
| eta | NDArray[np.float64] | Linear predictor including any offset, shape (n_rows,). | required |
| mu | NDArray[np.float64] | Response mean, shape (n_rows,). | required |
| response_residuals | NDArray[np.float64] | Observed response minus its mean, shape (n_rows,). | required |
| working_residuals | NDArray[np.float64] | Within working response minus design times coefficients, shape (n_rows,). | required |