Quick Start

This is the smallest UFXP path: fit the unnested structural estimator and read the same kind of reward, policy, and value objects as NFXP or CCP. The shortcut depends on empirical choice-probability support.

from econirl.datasets import load_rust_bus, rust_bus_reward_spec
from econirl import UFXP

df = load_rust_bus()

model = UFXP(n_states=90, discount=0.9999, utility=rust_bus_reward_spec(90))
model.fit(df, state="mileage_bin", action="replaced", id="bus_id")

print(model.params_)          # {"operating_cost": ..., "replacement_cost": ...}
print(model.se_)              # standard errors (optimal weighting)
print(model.summary())

Fitted attributes follow the same convention as NFXP and CCP:

Attribute	Meaning
`params_`	Estimated parameters by name.
`se_`	Standard errors from the efficient moment variance.
`coef_`	Coefficients as a numpy array.
`log_likelihood_`	Log-likelihood evaluated at the estimate.
`policy_`	Choice probabilities P(a given s), shape (n_states, n_actions).
`value_`	Value function V(s), shape (n_states,).
`converged_`	Whether the closed-form solve had full rank.

Weighting Modes

model = UFXP(n_states=90, weights="optimal")   # default: efficient, with SEs
model = UFXP(n_states=90, weights="random",    # plain random projections
             num_projections=64)               # consistent, no SEs

The default optimal weighting is the one to use; the random-projection mode is the paper’s baseline construction and is kept for comparison.

Full Estimator API

The underlying UFXPEstimator accepts a Panel and an explicit utility specification, problem definition, and transition tensor, mirroring the other low-level estimators:

from econirl.estimation import UFXPEstimator

result = UFXPEstimator().estimate(panel, utility, problem, transitions)