b34a1085165933fcffc6d863c2f7274b87bd2b09,qanta/reranker/svm.py,Reranker,create_train,#Any#Any#Any#,100

Before Change

    def create_train(example_generator, questions, guesses):
        data = dlib.sparse_ranking_pair()
        
        for row, gg in guesses.iterrows():
            guess = gg["guess"]
            answer = questions[gg["qnum"]].page
            text = questions[gg["qnum"]].get_text(gg["sentence"],
                                              gg["token"])

            if guess == answer:
                data.relevant.append(example_generator(text, gg[1]))
            else:
                data.nonrelevant.append(example_generator(text, gg[1]))
        return data

    def train_svm(data, c_val=10):
        trainer = dlib.svm_rank_trainer()

After Change

        queries = dlib.sparse_ranking_pairs()
        num_examples = 0

        for ex in Reranker.row_iterator(example_generator,
                                        questions, guesses):
            // Skip bad rows
            if ex.vector is None:
                continue

            if ex.new:
                if num_examples > 0 and has_correct:
                    queries.append(data)

                data = dlib.sparse_ranking_pair()
                has_correct = False

            if ex.is_correct:
                has_correct = True
                data.relevant.append(ex.vector)
            else:
                data.nonrelevant.append(ex.vector)
            num_examples += 1

        queries.append(data)
        return queries

    def score_one(self, example):

Instances