FlinkML is a library of sets of algorithms supported by Flink that can be used to solve real-life use cases. The algorithms are built so that they can use the distributed computing power of Flink and make predictions or do clustering and so on with ease. Right now, there are only a few sets of algorithms supported, but the list is growing.
FlinkML is being built with the focus on ML developers needing to write minimal glue code. Glue code is code that helps bind various components together. Another goal of FlinkML is to keep the use of algorithms simple.
Flink exploits in-memory data streaming and executes iterative data processing natively. FlinkML allows data scientists to test their models locally, with a subset of data, and then execute them in cluster mode on bigger data.
FlinkML is inspired by scikit-learn and Spark's MLlib, which allows you to define data pipelines cleanly and solve machine learning problems in a distributed manner.
The following is the road map Flink's development...