tfserving接口调用

http接口说明：

• 查看TensorFlow-Serving状态：
  curl http://localhost:8501/v1/models/${model_name}
• 查看TensorFlow-Serviing模型：
  curl http://localhost:8501/v1/models/${model_name}/metadata
• 使用Http请求进行模型预测：
  curl -d '{"instances": [1,2,3,4,5]}' -X POST http://localhost:8501/v1/models/${model_name}:predict
其中instances的value为模型输入Tensor的字符串形式，矩阵维度需要和Tensor对应。
• 查看模型情况
  saved_model_cli show --dir='./${model_path}/1' --all

python 调用restful接口：

def textclassification_cls():
    
    input_data_json = json.dumps({
        "instances": [{"user_sex":[1],
                       "patient_sex":[1],
                       "city_id":[5],
                       "user_age":[4],
                       "patient_age":[4],
                       "click":[0.002930],
                       "category_array":[1156,1156,1605,984],
                       "dep_ids":[112]+[0]*39,
                       "user_tag_zydjysnr90r": [31,3,9]+[0]*9,
                       "user_tag_public_department_click_30d": [7008, 10801,7009]+[0]*900,
                      "user_tag_disease_search_90d":[2799]+[0]*5
                    },{"user_sex":[1],
                       "patient_sex":[1],
                       "city_id":[5],
                       "user_age":[4],
                       "patient_age":[4],
                       "click":[0.002930],
                       "category_array":[1156,1156,1605,984],
                       "dep_ids":[112]+[0]*39,
                       "user_tag_zydjysnr90r": [31,3,9]+[0]*9,
                       "user_tag_public_department_click_30d": [7008, 10801,7009]+[0]*900,
                      "user_tag_disease_search_90d":[2799]+[0]*5
                    }]
    })

    SERVER_URL = 'http://10.1.2.106:8501/v1/models/deep-model:predict'
    # SERVER_URL = 'http://deepfm.api.91160.com/v1/models/deep-model:predict'
    import  time
    times = time.time()
    response = requests.post(SERVER_URL, data=input_data_json)
    res = json.loads(response.text)
    timee = time.time()
    print("耗时：",timee-times)
    return str(res)




if __name__ == "__main__":
    
    print(textclassification_cls())

存在的问题

1. 多维数组维度可以任意
2. padding 的0加不加结果一样