Realtime Advanced Analytics: Spark Streaming+Kafka, MLlib/GraphX, SQL/DataFrames
Details
Chris Freely, who recently left Databricks (Spark people) to join the IBM Spark Technology Center in San Francisco, will present a real-world, open source, advanced analytics and machine learning pipeline using all 20 Open Source technologies listed below.
This Meetup is based on Chris recent "Top-5" Hadoop Summit/Data Science talk called "Spark After Dark". Spark After Dark is a mock online dating site that uses Spark, Spark SQL, DataFrames, MLlib, GraphX, Cassandra, and ElasticSearch - among many other technologies listed below - to generate quality, real-time dating recommendations for its users.
Here are the Spark After Dark slides: http://www.slideshare.net/cfregly/spark-after-dark-real-time-advanced-analytics-and-machine-learning-with-spark
All code - and the entire pipeline runtime - will be dockerized and made publicly available on Github and the Docker Hub Registry.
Technologies to be demo'd:
- Apache Zeppelin (notebook-based development)
- Apache Spark SQL/DataFrames (Data Analysis and ETL)
- Apache Spark Streaming + Apache Kafka (Real-time Collection of Live Data from Interactive Demo)
- Spark Streaming + Real-time Machine Learning (K-Means Clustering, Log/Lin Regression)
- Apache Spark MLlib + GraphX (Generate personalized and non-personalized recommendations using various algorithms and feature engineering techniques including one hot encoding)
- MLlib + PMML Integration (Open Standard Markup Language for Predictive Models)
- Highly-scalable, NetflixOSS-based Machine Learning Prediction Serving Layer including Service Discover (Eureka) and Circuit Breakers (Hystrix) for Fault Tolerance
- Zeppelin + Python-based scikit-learn Machine Learning
- Spark + Neo4j = MazeRunner (Real-time Neo4j Graph Updates Beyond GraphX Batch Analytics)
- Spark R (Distributed R algorithmns)
- Apache Spark JDBC/ODBC Thrift Server (Beeline and Tableau Analytics Explorer Integration)
- Tachyon (Off-heap storage)
- Spark Job Server (REST API for managing Spark jobs)
- Spark + Cassandra (NoSQL, Lambda Arch Speed Layer)
- Spark + ElasticSearch (Distributed Search Engine)
- Spark + Redis (Distributed, Persistent Key-Value Store Similar to Memcached)
- Logstash (Log Agent + Collection)
- Kibana (ElasticSearch-based Analytics Explorer UI)
- HDFS + Parquet (Columnar Storage Format, Tight Compression, Lightning Fast Columnar Aggregations)
- Advanced visualizations within Zeppelin using python-based matplotlib and ggplot
Reminder that we'll be Docker-izing everything for you to reuse.
Keep an eye on the Github and Docker Hub Registry links under project name "fluxcapacitor":