Method process_tensor

python/paddle_serving_client/httpclient.py:399–472

(self, key, feed_dict, batch)

        return req

    def process_tensor(self, key, feed_dict, batch):
        lod = []
        if "{}.lod".format(key) in feed_dict:
            lod = feed_dict["{}.lod".format(key)]
        shape = self.feed_shapes_[key].copy()
        elem_type = self.feed_types_[key]
        data_value = feed_dict[key]
        data_key = proto_data_key_list[elem_type]
        proto_index = self.feed_names_to_idx_[key]
        name = self.feed_real_names[proto_index]
        alias_name = key

        # feed_dict[key] 可以是np.ndarray
        # 也可以是list或tuple
        # 当np.ndarray需要处理为list
        if isinstance(feed_dict[key], np.ndarray):
            shape_lst = []
            # 0维numpy 需要在外层再加一个[]
            if feed_dict[key].ndim == 0:
                data_value = [feed_dict[key].tolist()]
                shape_lst.append(1)
            else:
                shape_lst.extend(list(feed_dict[key].shape))
                shape = shape_lst
                data_value = feed_dict[key].flatten().tolist()
            # 当Batch为False，shape字段前插一个1,表示batch维
            # 当Batch为True,则直接使用numpy.shape作为batch维度
            if batch == False:
                shape.insert(0, 1)

            # 当是list或tuple时，需要把多层嵌套展开
        elif isinstance(feed_dict[key], (list, tuple)):
            # 当Batch为False，shape字段前插一个1,表示batch维
            # 当Batch为True, 由于list并不像numpy那样规整，所以
            # 无法获取shape，此时取第一维度作为Batch维度.
            # 插入到feedVar.shape前面.
            if batch == False:
                shape.insert(0, 1)
            else:
                shape.insert(0, len(feed_dict[key]))
            feed_dict[key] = [x for x in list_flatten(feed_dict[key])]
            data_value = feed_dict[key]
        else:
            # 输入可能是单个的str或int值等
            # 此时先统一处理为一个list
            # 由于输入比较特殊，shape保持原feedvar中不变
            data_value = []
            if isinstance(feed_dict[key], (str, bytes)):
                if self.feed_types_[key] != bytes_type:
                    raise ValueError(
                        "feedvar is not string-type,feed can`t be a single string."
                    )
                if isinstance(feed_dict[key], bytes):
                    feed_dict[key] = feed_dict[key].decode()
            else:
                if self.feed_types_[key] == bytes_type:
                    raise ValueError(
                        "feedvar is string-type,feed can`t be a single int or others."