Streamer is a complete software suite that enables data scientists to test ML algorithms on data streams. Streamer is open-source and multi-OS. It automates the entire data stream processing cycle and gives data scientists plenty of room to configure data streams and experiment in the most realistic conditions possible, all while minimizing setup time.
Streamer is an automatic data stream processing suite that allows data scientists to test continuous ML algorithms in realistic conditions. It handles every step of the processing cycle, from data collection to visualization of results.
Streamer also spares data scientists the complex and time-consuming task of hand-developing the processing cycle, so that they can concentrate on perfecting their algorithms. The suite offers preparation and post-processing tools, advanced ML algorithms for streams, and evaluation tools. It also offers APIs that enable integration of third-party tools and algorithms written in a variety of programming languages, like Python, R, and Java. It also features a GUI that simplifies the monitoring of ML and data stream analysis processes.
Streamer’s code is open source (GNU GPL 3 license), allowing data scientists to modify it and add any desired features. However, the suite is also ready to use as-is in operational contexts.
Streamer was developed through collaboration between CEA-List and DAVID (a lab at Paris-Saclay University focused on sustainable, data-driven smart city projects) as part of the StreamOps project, financed by DATAIA.
We use Streamer, which we co-developed with CEA-List, in our ANR project Polluscope. The project aims to develop algorithms that can describe individuals’ air pollution exposure by leveraging a stream of measurement data collected by microsensors.
Streamer facilitates the development of AI solutions implementing continuous learning.
Cybersecurity experts increasingly count on ML algorithms for the evaluation and contextualization of threats and alerts. As part of an in-house research program, CEA-List demonstrated the advantages that Streamer brings to the creation and use of these algorithms:
STREAMER: a Powerful and Open-Source Framework for Continuous Learning in Data Streams, Garcia-Rodriguez, Sandra, Mohammad Alshaer, and Cedric Gouy-Pailler. Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 2020.
Detecting Anomalies from Streaming Time Series Using Matrix Profile and Shapelets Learning, Mohammad Alshaer, Sandra Garcia-Rodriguez and Cedric Gouy-Pailler. Proceedings of the 32th ACM International Conference on Tools with Artificial Intelligence. 2020.
Learn more: Streamer