On the strong consistency of approximate maximum likelihood estimators