Quick Start

This page shows the public TD-CCP wrapper. The snippet is about using the API; the reason to choose TD-CCP is the transition-density-free estimation route described on the parent page.

The public API follows the same convention as the other structural estimators: create an estimator, call fit, and inspect fitted attributes.

from econirl.datasets import load_rust_bus
from econirl import TDCCP

df = load_rust_bus()

model = TDCCP(
    n_states=90,
    n_actions=2,
    discount=0.9999,
    utility="linear_cost",
    method="semigradient",
)
model.fit(df, state="mileage_bin", action="replaced", id="bus_id")

print(model.params_)
print(model.se_)
print(model.policy_.shape)

The fitted estimator exposes the estimated reward parameters, standard errors, policy, value function, and likelihood diagnostics.

Attribute	Meaning
`params_`	Estimated reward parameters
`se_`	Standard errors for those parameters
`policy_`	Estimated action probabilities by state
`value_`	Estimated value function by state
`log_likelihood_`	Maximized CCP pseudo log likelihood
`ev_features_`	Continuation-value feature decomposition when available

Method Choices

semigradient = TDCCP(method="semigradient", basis_type="polynomial")
encoded = TDCCP(method="semigradient", basis_type="encoded")
neural = TDCCP(method="neural", avi_iterations=20)

Use the semigradient path for the current reported workflow. Use the encoded basis when states already have meaningful numeric encoders. The neural AVI path is available for flexible approximation, but it is not the current simulation-study target.

Full Estimator API

Use econirl.estimation.td_ccp.TDCCPEstimator when you need direct control over panels, utility objects, basis settings, cross-fitting, robust standard errors, or supplied transition tensors for policy and value evaluation.