StreamSets Transformer
Extensibility — Part 2:
Spark MLeap Bundles to S3

StreamSets Transformer
Extensibility — Part 2:
Spark MLeap Bundles to S3

In part 1, you learned how to extend StreamSets Transformer in order to train Spark ML RandomForestRegressor model.

In this part 2, you will learn how to create Spark MLeap bundle to serialize the trained model and save the bundle to Amazon S3.

MLeap is a common serialization format and execution engine for machine learning pipelines. It supports Spark, Scikit-learn and Tensorflow for training pipelines and exporting them to an MLeap Bundle. Serialized pipelines (bundles) can be deserialized back into Spark for batch-mode scoring or the MLeap runtime to power realtime API services.

For more details on MLeap, click here.

StreamSets Transformer Pipeline Overview

Before we dive into code, here is the extended pipeline overview.

Prerequisites: For this pipeline to run, these external libraries must be installed from the UI via Package Manager under Basic Stage Library. (Note that this will require Transformer restart.)

Scala Processor

Below is the additional Scala code that takes the trained model, creates an MLeap bundle, and stores that in a temporary location as defined by saveModelZipPath variable.

For details on input data, preparing features and other such details, refer to part 1.

Here's the entire code snippet that's inserted into Scala processor >> Scala >> Scala Code section. The code of interest is from lines 61 to 74.

Note: Under the covers, this custom code is compiled into a jar and handed-off to the underlying execution engine — which in this case is Spark.

Stream Selector Processor

In this processor we're checking to see if our accuracy “threshold” is met by comparing the value of R2 that was calculated after training the model and running inference on “test” set.

Note: Similarly, other condition(s) can be applied to any metric calculated and made available during model training process.

Scala Processor

If the accuracy “threshold” of R2 > 0.75 is met, the pipeline execution will continue down this path and the following Scala code will get executed. The code snippet takes the MLeap bundle (“model.zip”) created during the model training process and uploads it to Amazon S3.

Here's the Scala code inserted into Scala processor >> Scala >> Scala Code section.

Output

Assuming things go well without any errors, running the pipeline will result in serialized MLeap bundle (“model.zip”) getting uploaded to Amazon S3 as shown below.

Note that downloading the serialized MLeap bundle (“model.zip”) from Amazon S3 should reveal trained Spark ML model's contents similar to as shown below.

Stay tuned!

In this blog you learned how easily you can extend StreamSets Transformer’s functionality. In particular, you learned how to incorporate custom Scala code to train the Spark ML machine learning model, create a serialized MLeap bundle for it, and upload that to Amazon S3.

In part 3, I will illustrate how to load the serialized MLeap bundle from Amazon S3 and use the model for prediction on new data flowing through StreamSets Transformer pipeline. Stay tuned!

To learn more about StreamSets Transformer visit our website, refer to the documentation, join the webinar, and download the binaries.

Related Resources

Check out StreamSets white papers, videos, webinars, report and more.

Visit the Resource Library

Related Blog Posts

Receive Updates

Receive Updates

Join our mailing list to receive the latest news from StreamSets.

You have Successfully Subscribed!