def _serialize_example()

in src/data_preprocess/tf_records_generator.py [0:0]

• 35 lines of code
• 14 McCabe index (conditional complexity)

    def _serialize_example(self, obj_id, sequence, obj, label):
        d_feature = {}
        d_feature['id'] = self._bytes_feature(obj_id.encode())
        d_feature['seq'] = self._bytes_feature(sequence.encode())
        d_feature['L'] = self._int_feature(len(sequence))
        cur_example_label = label
        if isinstance(cur_example_label, dict):
            assert self.label_type is not None
            cur_example_label = cur_example_label[self.label_type]
        if cur_example_label is None or len(cur_example_label) == 0:
            return None
        if isinstance(cur_example_label, list) or isinstance(cur_example_label, set):
            self.labels = self.labels.union(set(cur_example_label))
            cur_example_label_ids = [self.label_2_id[v] for v in cur_example_label]
            d_feature['label'] = self._int_feature(cur_example_label_ids)
        else:
            self.labels.add(cur_example_label)
            cur_example_label_id = self.label_2_id[cur_example_label]
            d_feature['label'] = self._int_feature(cur_example_label_id)

        for item in obj.items():
            name = item[0]
            dtype = item[1][0]
            value = item[1][1]
            if isinstance(value, np.ndarray):
                value = list(value.reshape(-1))
            elif isinstance(value, int) or isinstance(value, float) or isinstance(value, str):
                value = [value]
            if dtype == "str":
                d_feature[name] = self._bytes_feature(value)
            elif dtype == "int":
                d_feature[name] = self._int_feature(value)
            else:
                d_feature[name] = self._float_feature(value)

        example = tf.train.Example(features=tf.train.Features(feature=d_feature))
        return example.SerializeToString()