ed3a343d536daeb51eb0e39e630717c32c611766,abcpy/methods.py,Rejection,infer,#Rejection#Any#Any#,42

Before Change

        

        // only run at first call unless quantile specified
        if not hasattr(self, "distances") or quantile is not None:
            self.distances = np.empty((0,1))
            self.parameters = [ np.empty((0,1)) for ii in range(self.n_params) ]

            if quantile is not None:
                distances, parameters = self._get_distances(int(self.n_samples / quantile))
                threshold = np.percentile(distances, quantile*100)
                discard_rest = True
                save_values = True

            else:
                distances, parameters = self._get_distances(self.n_samples)
                discard_rest = False
                save_values = True

        else:  // use precomputed distances
            distances = self.distances
            parameters = self.parameters
            discard_rest = False
            save_values = False

        posteriors = self._apply_threshold(distances, parameters, threshold, discard_rest, save_values)

        return {"samples": posteriors, "threshold": threshold}

After Change

            threshold = np.percentile(distances, quantile*100)

        // filter too dissimilar samples
        accepted = distances < threshold
        posteriors = [p[accepted] for p in parameters]

        return {"samples": posteriors, "threshold": threshold}

Instances