logit_rules.py 文件源码

def flat_prior(self, state):
        """ Evaluate log-probability of each primitive in the population.

        Return value is properly normalized.

        """
        raw_phylogeny_weights = self.rule_population.flat_variable_weights 
        phylogeny_weights = scipy.stats.norm.logpdf(
            np.log(raw_phylogeny_weights),
            loc = state.phylogeny_mean,
            scale = state.phylogeny_std
        )

        total_duration = float(self.data.experiment_end - self.data.experiment_start)
        durations = (self.rule_population.flat_durations /
                     ((1.0+self.window_duration_epsilon)*total_duration)
                    )
        window_a = (
            state.window_concentration *
            state.window_typical_fraction
        )
        window_b = (
            state.window_concentration *
            (1.0-state.window_typical_fraction)
        )
        window_weights = scipy.stats.beta.logpdf(
            durations, 
            window_a,
            window_b
        )

        weights = phylogeny_weights + window_weights
        normalization = logsumexp(weights)
        return weights - normalization