This Python app generates data, analyzes it in Spark Streaming, and visualizes the results with Lightning. The analyses use streaming machine learning algorithms included with Spark as of version 1.2. The demos are designed for local use, but the same algorithms can run at scale on a cluster with millions of records.
To run these demos, you need:
- A working installation of Spark
- A running Lightning server
- An installation of Python with standard scientific computing libraries (NumPy, SciPy, ScikitLearn)
With those three things in place, install using:
pip install spark-ml-streaming
Then set SPARK_HOME to your Spark installation, and start an executable:
streaming-kmeans -l <lighting_host>
Where lightning_host is the address of your Lightning server. After it starts, your browser will open, and you should see data appear shortly.
Try running with different settings, for example, to run a 1-d version with 4 clusters and a half-life of 10 points:
streaming-kmeans -p <temporary_path> -l <lighting_host> -nc 4 -nd 1 -hl 10 -tu points
Where temporary_path is where data will be written / read, if not specified the current tmp directory will be used (See Python tempfile.gettempdir())
2D data will make a scatter plot and 1D data will make a line plot. You can set this with -nd.
To see all options type:
streaming-kmeans -h
The demo relies on a Scala package included pre-built inside python/mlstreaming/lib. To rebuild it, use sbt:
cd scala
sbt package