7. Data Science

Register the out-of-the-box applications for the Kafka binder

Note

These samples assume that the Data Flow Server can access a remote Maven repository, repo.spring.io/libs-release by default. If your Data Flow server is running behind a firewall, or you are using a maven proxy preventing access to public repositories, you will need to install the sample apps in your internal Maven repository and configure the server accordingly. The sample applications are typically registered using Data Flow’s bulk import facility. For example, the Shell command dataflow:>app import --uri dataflow.spring.io/rabbitmq-maven-latest (The actual URI is release and binder specific so refer to the sample instructions for the actual URL). The bulk import URI references a plain text file containing entries for all of the publicly available Spring Cloud Stream and Task applications published to repo.spring.io. For example, source.http=maven://org.springframework.cloud.stream.app:http-source-rabbit:2.1.0.RELEASE registers the http source app at the corresponding Maven address, relative to the remote repository(ies) configured for the Data Flow server. The format is maven://<groupId>:<artifactId>:<version> You will need to download the required apps or build them and then install them in your Maven repository, using whatever group, artifact, and version you choose. If you do this, register individual apps using dataflow:>app register…​ using the maven:// resource URI format corresponding to your installed app.

dataflow:>app import --uri https://dataflow.spring.io/kafka-maven-latest

Create and deploy the following stream

dataflow:>stream create --name pmmlTest --definition "http --server.port=9001 | pmml --modelLocation=https://raw.githubusercontent.com/spring-cloud/spring-cloud-stream-modules/master/pmml-processor/src/test/resources/iris-flower-classification-naive-bayes-1.pmml.xml --inputs='Sepal.Length=payload.sepalLength,Sepal.Width=payload.sepalWidth,Petal.Length=payload.petalLength,Petal.Width=payload.petalWidth' --outputs='Predicted_Species=payload.predictedSpecies' --inputType='application/x-spring-tuple' --outputType='application/json'| log" --deploy
Created and deployed new stream 'pmmlTest'

Note

The built-in pmml processor will load the given PMML model definition and create an internal object representation that can be evaluated quickly. When the stream receives the data, it will be used as the input for the evaluation of the analytical model iris-flower-classifier-1 contained in the PMML document. The result of this evaluation is a new field predictedSpecies that was created from the pmml processor by applying a classifier that uses the naiveBayes algorithm.

Verify the stream is successfully deployed

dataflow:>stream list

Notice that pmmlTest.http, pmmlTest.pmml, and pmmlTest.log Spring Cloud Stream applications are running within the local-server.

2016-02-18 06:36:45.396  INFO 31194 --- [nio-9393-exec-1] o.s.c.d.d.l.OutOfProcessModuleDeployer   : deploying module org.springframework.cloud.stream.module:log-sink:jar:exec:1.0.0.BUILD-SNAPSHOT instance 0
   Logs will be in /var/folders/c3/ctx7_rns6x30tq7rb76wzqwr0000gp/T/spring-cloud-data-flow-3038434123335455382/pmmlTest-1455806205386/pmmlTest.log
2016-02-18 06:36:45.402  INFO 31194 --- [nio-9393-exec-1] o.s.c.d.d.l.OutOfProcessModuleDeployer   : deploying module org.springframework.cloud.stream.module:pmml-processor:jar:exec:1.0.0.BUILD-SNAPSHOT instance 0
   Logs will be in /var/folders/c3/ctx7_rns6x30tq7rb76wzqwr0000gp/T/spring-cloud-data-flow-3038434123335455382/pmmlTest-1455806205386/pmmlTest.pmml
2016-02-18 06:36:45.407  INFO 31194 --- [nio-9393-exec-1] o.s.c.d.d.l.OutOfProcessModuleDeployer   : deploying module org.springframework.cloud.stream.module:http-source:jar:exec:1.0.0.BUILD-SNAPSHOT instance 0
   Logs will be in /var/folders/c3/ctx7_rns6x30tq7rb76wzqwr0000gp/T/spring-cloud-data-flow-3038434123335455382/pmmlTest-1455806205386/pmmlTest.http

Post sample data to the http endpoint: localhost:9001 (9001 is the port we specified for the http source in this case)

dataflow:>http post --target http://localhost:9001 --contentType application/json --data "{ \"sepalLength\": 6.4, \"sepalWidth\": 3.2, \"petalLength\":4.5, \"petalWidth\":1.5 }"
> POST (application/json;charset=UTF-8) http://localhost:9001 { "sepalLength": 6.4, "sepalWidth": 3.2, "petalLength":4.5, "petalWidth":1.5 }
> 202 ACCEPTED

Verify the predicted outcome by tailing <PATH/TO/LOGAPP/pmmlTest.log/stdout_0.log file. The predictedSpecies in this case is versicolor.

{
  "sepalLength": 6.4,
  "sepalWidth": 3.2,
  "petalLength": 4.5,
  "petalWidth": 1.5,
  "Species": {
    "result": "versicolor",
    "type": "PROBABILITY",
    "categoryValues": [
      "setosa",
      "versicolor",
      "virginica"
    ]
  },
  "predictedSpecies": "versicolor",
  "Probability_setosa": 4.728207706362856E-9,
  "Probability_versicolor": 0.9133639504608079,
  "Probability_virginica": 0.0866360448109845
}

Let’s post with a slight variation in data.

dataflow:>http post --target http://localhost:9001 --contentType application/json --data "{ \"sepalLength\": 6.4, \"sepalWidth\": 3.2, \"petalLength\":4.5, \"petalWidth\":1.8 }"
> POST (application/json;charset=UTF-8) http://localhost:9001 { "sepalLength": 6.4, "sepalWidth": 3.2, "petalLength":4.5, "petalWidth":1.8 }
> 202 ACCEPTED

	Note
	petalWidth value changed from 1.5 to 1.8

The predictedSpecies will now be listed as virginica.

{
  "sepalLength": 6.4,
  "sepalWidth": 3.2,
  "petalLength": 4.5,
  "petalWidth": 1.8,
  "Species": {
    "result": "virginica",
    "type": "PROBABILITY",
    "categoryValues": [
      "setosa",
      "versicolor",
      "virginica"
    ]
  },
  "predictedSpecies": "virginica",
  "Probability_setosa": 1.0443898084700813E-8,
  "Probability_versicolor": 0.1750120333571921,
  "Probability_virginica": 0.8249879561989097
}