排序模型在线测试

日萌社

人工智能AI：Keras PyTorch MXNet TensorFlow PaddlePaddle 深度学习实战（不定时更新）

---

7. 8 排序模型在线测试

学习目标

• 目标
  • 无
• 应用
  • 应用TensorFlow Serving apis完成在线模型的获取排序测试

7.8.1 排序模型在线预测添加

• 目的：编写tf serving客户端程序调用serving模型服务，进行在线预测测试
• 步骤：
  • 1、用户特征与文章特征合并
  • 2、serving服务端的example样本结构构造
  • 3、模型服务调用对接

1、用户特征与文章特征合并

import tensorflow as tf
from grpc.beta import implementations
from tensorflow_serving.apis import prediction_service_pb2_grpc
from tensorflow_serving.apis import predict_pb2
from tensorflow_serving.apis import classification_pb2
import os
import sys
import grpc
from server.utils import HBaseUtils
from server import pool
import numpy as np
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
print(BASE_DIR)
sys.path.insert(0, os.path.join(BASE_DIR))


def wdl_sort_service():
    """
    wide&deep进行排序预测
    :param reco_set:
    :param temp:
    :param hbu:
    :return:
    """
    hbu = HBaseUtils(pool)
    # 排序
    # 1、读取用户特征中心特征
    try:
        user_feature = eval(hbu.get_table_row('ctr_feature_user',
                                              '{}'.format(1115629498121846784).encode(),
                                              'channel:{}'.format(18).encode()))
        # logger.info("{} INFO get user user_id:{} channel:{} profile data".format(
        #     datetime.now().strftime('%Y-%m-%d %H:%M:%S'), temp.user_id, temp.channel_id))
    except Exception as e:
        user_feature = []
    if user_feature:
        # 2、读取文章特征中心特征
        result = []

        # examples
        examples = []
        for article_id in [17749, 17748, 44371, 44368]:
            try:
                article_feature = eval(hbu.get_table_row('ctr_feature_article',
                                                         '{}'.format(article_id).encode(),
                                                         'article:{}'.format(article_id).encode()))
            except Exception as e:

                article_feature = [0.0] * 111

2、serving服务端的example样本结构构造

channel_id = int(article_feature[0])
# 求出后面若干向量的平均值
vector = np.mean(article_feature[11:])
# 第三个用户权重特征
user_feature = np.mean(user_feature)
# 第四个文章权重特征
article_feature = np.mean(article_feature[1:11])

# 组建example
example = tf.train.Example(features=tf.train.Features(feature={
    "channel_id": tf.train.Feature(int64_list=tf.train.Int64List(value=[channel_id])),
    "vector": tf.train.Feature(float_list=tf.train.FloatList(value=[vector])),
    'user_weights': tf.train.Feature(float_list=tf.train.FloatList(value=[user_feature])),
    'article_weights': tf.train.Feature(float_list=tf.train.FloatList(value=[article_feature])),
}))

examples.append(example)

3、模型服务调用对接

使用 grpc.insecure_channel('localhost:port') 进行连接 服务端， 接着在这个 channel 上创建 stub

• 导入包：
  • from tensorflow_serving.apis import prediction_service_pb2_grpc：与grpc建立stub
    • PredictionServiceStub：建立stub，负责将调用的接口、方法和参数约定
      • Classify:发送请求, 设置延时时间 secs timeout
  • from tensorflow_serving.apis import classification_pb2：调用服务，提供请求参数

with grpc.insecure_channel('127.0.0.1:8500') as channel:
    stub = prediction_service_pb2_grpc.PredictionServiceStub(channel)

    # 获取测试数据集，并转换成 Example 实例
    # 准备 RPC 请求，指定模型名称。
    request = classification_pb2.ClassificationRequest()
    request.model_spec.name = 'wdl'
    request.input.example_list.examples.extend(examples)

    # 获取结果
    response = stub.Classify(request, 10.0)
    print(response)

4、在线排序代码预测模型服务对接测试

这里我们单独对这个WDL模型进行测试，直接在sort_serveice中增加这个wide&deep的函数。然后直接本函数内测试(不加入到实时推荐中测试)。注意测试前注释掉server的__init.py文件中之前创建spark session的初始化函数(因为pycharm运行需要配置环境)

# 排序模型所需要的spark
# from pyspark import SparkConf
# from pyspark.sql import SparkSession
# # spark配置
# conf = SparkConf()
# conf.setAll(DefaultConfig.SPARK_GRPC_CONFIG)
# 
# SORT_SPARK = SparkSession.builder.config(conf=conf).getOrCreate()

运行测试结果：

result {
  classifications {
    classes {
      label: "0"
      score: 0.9064760208129883
    }
    classes {
      label: "1"
      score: 0.09352399408817291
    }
  }
  classifications {
    classes {
      label: "0"
      score: 0.9092260599136353
    }
    classes {
      label: "1"
      score: 0.09077390283346176
    }
  }
  classifications {
    classes {
      label: "0"
      score: 0.9146808981895447
    }
    classes {
      label: "1"
      score: 0.08531906455755234
    }
  }
  classifications {
    classes {
      label: "0"
      score: 0.9088123440742493
    }
    classes {
      label: "1"
      score: 0.09118763357400894
    }
  }
}
model_spec {
  name: "wdl"
  version {
    value: 1561990083
  }
  signature_name: "serving_default"
}

完整代码：

import tensorflow as tf
from grpc.beta import implementations
from tensorflow_serving.apis import prediction_service_pb2_grpc
from tensorflow_serving.apis import predict_pb2
from tensorflow_serving.apis import classification_pb2
import os
import sys
import grpc
from server.utils import HBaseUtils
from server import pool
import numpy as np
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
print(BASE_DIR)
sys.path.insert(0, os.path.join(BASE_DIR))


def wdl_sort_service():
    """
    wide&deep进行排序预测
    :param reco_set:
    :param temp:
    :param hbu:
    :return:
    """
    hbu = HBaseUtils(pool)
    # 排序
    # 1、读取用户特征中心特征
    try:
        user_feature = eval(hbu.get_table_row('ctr_feature_user',
                                              '{}'.format(1115629498121846784).encode(),
                                              'channel:{}'.format(18).encode()))
        # logger.info("{} INFO get user user_id:{} channel:{} profile data".format(
        #     datetime.now().strftime('%Y-%m-%d %H:%M:%S'), temp.user_id, temp.channel_id))
    except Exception as e:
        user_feature = []
    if user_feature:
        # 2、读取文章特征中心特征
        result = []

        # examples
        examples = []
        for article_id in [17749, 17748, 44371, 44368]:
            try:
                article_feature = eval(hbu.get_table_row('ctr_feature_article',
                                                         '{}'.format(article_id).encode(),
                                                         'article:{}'.format(article_id).encode()))
            except Exception as e:

                article_feature = [0.0] * 111

            channel_id = int(article_feature[0])
            # 求出后面若干向量的平均值
            vector = np.mean(article_feature[11:])
            # 第三个用户权重特征
            user_feature = np.mean(user_feature)
            # 第四个文章权重特征
            article_feature = np.mean(article_feature[1:11])

            # 组建example
            example = tf.train.Example(features=tf.train.Features(feature={
                "channel_id": tf.train.Feature(int64_list=tf.train.Int64List(value=[channel_id])),
                "vector": tf.train.Feature(float_list=tf.train.FloatList(value=[vector])),
                'user_weights': tf.train.Feature(float_list=tf.train.FloatList(value=[user_feature])),
                'article_weights': tf.train.Feature(float_list=tf.train.FloatList(value=[article_feature])),
            }))

            examples.append(example)

        with grpc.insecure_channel('127.0.0.1:8500') as channel:
            stub = prediction_service_pb2_grpc.PredictionServiceStub(channel)

            # 获取测试数据集，并转换成 Example 实例
            # 准备 RPC 请求，指定模型名称。
            request = classification_pb2.ClassificationRequest()
            request.model_spec.name = 'wdl'
            request.input.example_list.examples.extend(examples)

            # 获取结果
            response = stub.Classify(request, 10.0)
            print(response)

    return None


if __name__ == '__main__':
    wdl_sort_service()