线性回归模型的实时开放坏要求

Question

采用线性回归方法，从"http://datascience-enthusiast.com“出发，根据温度、真空度、压力和湿度的变化值，进行功率预测，并将模型应用到卡夫卡主题的实时数据中。正确生成腌制的.pkl.z文件，并按照建议的https://github.com/jpmml/jpmml-sklearn使用JPMML将其转换为PMML。

program (kafka_producer.py)在某些范围内随机生成浮点数数据，并将字符串以字节的形式发送到Kafka主题。

作为Openscoring的python消费程序(kafka_consumer.py)从Kafka主题中读取数据，将字节字符串转换为字符串，最后转换为一个字典，该字典形成参数= {"AT“：9.2、"V”：39.82、"AP“：1013.19、"RH”：91.25}表示结果= os.evaluate("CCPP"，参数)语句。

它工作良好，预测能力，但在4至10条记录中正确显示结果后，Openscoring服务器抛出。

SEVERE: INFO:
Received EvaluationRequest{id=null, arguments={AT=12.12, V=41.35, AP=1031.67, RH=66.32}}
Nov 20, 2017 6:39:16 AM org.openscoring.service.ModelResource evaluate
INFO: Returned EvaluationResponse{id=null, result={PE=472.152110955029}}
Nov 20, 2017 6:39:17 AM org.openscoring.service.ModelResource evaluate
INFO: Received EvaluationRequest{id=null, arguments={AT=34.06, V=51.53, AP=1016.22, RH=91.7}}
Nov 20, 2017 6:39:17 AM org.openscoring.service.ModelResource evaluate
INFO: Returned EvaluationResponse{id=null, result={PE=444.9147880324237}}
Nov 20, 2017 6:39:18 AM org.openscoring.service.ModelResource evaluate
INFO: Received EvaluationRequest{id=null, arguments={AT=20.41, V=50.33, AP=1018.19, RH=100.18}}
Nov 20, 2017 6:39:18 AM org.openscoring.service.ModelResource doEvaluate
**SEVERE: Failed to evaluate**
org.jpmml.evaluator.InvalidResultException (at or around line 130)
at org.jpmml.evaluator.FieldValueUtil.performInvalidValueTreatment(FieldValueUtil.java:178) at org.jpmml.evaluator.FieldValueUtil.prepareInputValue(FieldValueUtil.java:90)
at org.jpmml.evaluator.InputField.prepare(InputField.java:64)

卡夫卡消费者停止并显示:引发异常(self.message)异常:坏请求

kafka_producer.py
import random
import time
from kafka import KafkaProducer
from kafka.errors import KafkaError
producer = KafkaProducer(bootstrap_servers='localhost:9092')
topic = "power"
for i in range(1000):
AT = "19.651231"
V = "54.305804"
AP = "1013.259078"
RH = "73.308978"
def getAT():
return str(round(random.uniform(2.0, 38.0),2))
def getV():
return str(round(random.uniform(26.0, 81.5),2))
def getAP():
return str(round(random.uniform(993.0, 1033.0),2))
def getRH():
return str(round(random.uniform(26.0, 101.0),2))
# arguments = {"AT" :9.2, "V" : 39.82, "AP" : 1013.19, "RH" : 91.25}
message = "{"AT" : " + getAT() + "," + ""V" : " +getV() + "," + ""AP" : " +getAP() + "," + ""RH" : " + getRH() + "}"
producer.send(topic, key=str.encode('key_{}'.format(i)), value=(message.encode('utf-8')))
time.sleep(1)
producer.close()

kafka_consumer.py
import ast
from kafka import KafkaConsumer
import openscoring
import os
os = openscoring.Openscoring("http://localhost:8080/openscoring")
kwargs = {"auth" : ("admin", "adminadmin")}
os.deploy("CCPP", "/home/gopinathankm/jpmml-sklearn-master/ccpp.pmml", **kwargs)
consumer = KafkaConsumer('power', bootstrap_servers='localhost:9092')
for message in consumer:
arguments =message.value
argsdict = arguments.decode("utf-8")
dict = ast.literal_eval(argsdict)
print(dict)
result = os.evaluate("CCPP", dict)
print(result)

对于一些生成的数据，它不起作用，我真的不知道生成的请求有多糟糕。如能提供任何协助，将不胜感激。Gopinathan K.M .

Answer 1

从Villu Ruusmann获得了以下援助和解决办法，以便其他人受益：

https://github.com/jpmml/jpmml-evaluator/issues/84

“异常类型InvalidResultException意味着无法成功地完成模型的计算，因为一个或多个输入字段值超出了它们声明的范围。您的PMML文档和Python的值范围不匹配。要么删除PMML文档值范围(这样所有输入值都被认为是有效的)，要么约定Python值范围。”如维卢·鲁斯曼所言。