econirl.SufficientStats

class econirl.SufficientStats(state_action_counts, transitions, empirical_ccps, initial_distribution, n_observations, n_individuals)[source]

Bases: object

Pre-computed statistics for tabular estimators.

Tabular estimators (NFXP, CCP) only need these summary statistics, not individual trajectories. This makes estimation O(1) in dataset size.

Variables:

  • state_action_counts (jax.jaxlib._jax.Array) – Array of shape (S, A) with raw observation counts.

  • transitions (jax.jaxlib._jax.Array) – Array of shape (A, S, S) with estimated P(s'|s,a).

  • empirical_ccps (jax.jaxlib._jax.Array) – Array of shape (S, A) with empirical P(a|s).

  • initial_distribution (jax.jaxlib._jax.Array) – Array of shape (S,) with empirical starting state distribution across individuals.

  • n_observations (int) – Total number of (s, a, s’) observations.

  • n_individuals (int) – Number of distinct individuals in the panel.

Parameters:

  • state_action_counts (Array)

  • transitions (Array)

  • empirical_ccps (Array)

  • initial_distribution (Array)

  • n_observations (int)

  • n_individuals (int)

state_action_counts: Array
transitions: Array
empirical_ccps: Array
initial_distribution: Array
n_observations: int
n_individuals: int
property n_states: int

Number of states.

property n_actions: int

Number of actions.

__init__(state_action_counts, transitions, empirical_ccps, initial_distribution, n_observations, n_individuals)
Parameters:

  • state_action_counts (Array)

  • transitions (Array)

  • empirical_ccps (Array)

  • initial_distribution (Array)

  • n_observations (int)

  • n_individuals (int)

Return type:

None