Pre-Estimation Checks

Read this page before fitting MCE-IRL. The checks tell you whether the supplied features, transitions, and normalization can support reward recovery from demonstrations.

Most MCE-IRL failures are visible before optimization starts. Check feature variation, support, transition validity, and normalization before treating a fit as reward recovery.

Check	Why it matters
Feature rank	Rank below the number of reward features means some reward directions are unidentified.
Feature condition number	A high condition number signals unstable feature matching.
Action dependence	Multi-action reward recovery needs features that vary across actions.
Transition row sums	Occupancy measures require valid probability rows.
State coverage	Unobserved states weaken the occupancy comparison.
Action support	Rare actions make action-specific rewards weakly pinned down.
Reward anchor	MCE rewards need a normalization for level and scale.

Canonical Simulation Checks

Values from the primary mce_low_high_reward synthetic run (see Simulation Study):

Check	Value	Status
Feature rank	8 / 8	pass
Feature condition number	1.373	pass
Action-dependent features	true	pass
Transition row error	1.86e-8	pass
Observed states	25 / 25	pass
State-action coverage	1.000	pass
Action shares	0.334, 0.336, 0.330	pass
Minimum positive CCP	0.130	pass
Reward anchor	true	pass

Common Risk Patterns

A state-only feature matrix can produce a plausible reward vector while still failing to identify action-specific payoffs. Sparse demonstrations can match the dominant actions while leaving rare-action rewards weak. Transition tensors with the wrong orientation can have valid row sums and still produce the wrong occupancy measure.

The wrapper checks the action-contrast feature rank at fit time and warns when it is deficient. A fit that cannot identify action-specific payoffs is flagged before the estimates are read.