model_entailment.py 文件源码

def process_test_data(self, input_file, onto_aware, is_labeled=True):
        if not self.model:
            raise RuntimeError, "Model not trained yet!"
        print >>sys.stderr, "Reading test data"
        label_ind = []
        tagged_sentences = []
        for line in open(input_file):
            lnstrp = line.strip()
            if is_labeled:
                label, tagged_sentence = lnstrp.split("\t")
                if label not in self.label_map:
                    self.label_map[label] = len(self.label_map)
                label_ind.append(self.label_map[label])
            else:
                tagged_sentence = lnstrp
            tagged_sentences.append(tagged_sentence)
        print >>sys.stderr, "Indexing test data"
        # Infer max sentence length if the model is trained
        input_shape = self.model.get_input_shape_at(0)[0]  # take the shape of the first of two inputs at 0.
        sentlenlimit = input_shape[1]  # (num_sentences, num_words, num_senses, num_hyps)
        test_inputs = self.data_processor.prepare_paired_input(tagged_sentences, onto_aware=onto_aware,
                                                               sentlenlimit=sentlenlimit, for_test=True)
        test_labels = self.data_processor.make_one_hot(label_ind)
        return test_inputs, test_labels