e81142f0869321cd8792e13bfc4f6fb85d9f1efc,libact/query_strategies/uncertainty_sampling.py,UncertaintySampling,make_query,#UncertaintySampling#,62

Before Change

            // therefore, total time complexity is O(NK) + O(NK) + O(NK) + O(N) = O(NK)
            ask_id = np.argmax(1 - np.max(self.model.predict_real(X_pool), 1))

        elif self.method == "sm":  // smallest margin
            // time complexity analysis:
            // O(NK) + O(N)
            prob = self.model.predict_real(X_pool)
            min_margin = np.inf
            for j in range(len(prob)):
                m1_id = np.argmax(prob[j])
                m2_id = np.argmax(np.delete(prob[j], m1_id))
                margin = prob[j][m1_id] - prob[j][m2_id]
                if margin < min_margin:
                    min_margin = margin
                    ask_id = j

        else:
            raise ValueError(
                "Invalid method "%s" (available choices: ("lc", "sm", "le")"
                % self.method
                )

        return unlabeled_entry_ids[ask_id]

    def get_model(self):

After Change

            else:
                // Find 2 largest decision values
                ind = np.partition(-np.abs(dvalue), 2, axis=1)[:2]
            margin = np.abs(np.abs(dvalue[:, ind[0]]) - np.abs(dvalue[:, ind[1]]))
            ask_id = np.argmin(margin)

        return unlabeled_entry_ids[ask_id]

Instances