As we are continually looking for ways to improve and enhance Netdata, we are starting to explore how we can leverage machine learning to introduce new product features.
Our main focus at the moment is around automated anomaly detection. This is a really interesting and challenging problem (high volume, high dimensional data, lack of ground truth labels, and so on), but we should be able to use some of the metrics monitored by Netdata to deliver new, awesome product features and user experiences (AI is the new electricity, after all 😃).
However, developing ML-driven product features is quite different than traditional software development (see steps 1 to 7 in the picture above). Mainly, this is because you never really know what specific data transformations, problem formulation, and sets of algorithms will work best in advance. (Here is a good article explaining things, and if you really want to go down a rabbit hole, check out this Stack Overflow question and this Quora thread).
Ideally, you first need to prototype your solution “in the lab” on some data you have already collected and do a few iterations of data → problem formulation → prototype. This process gives you a level of confidence in what you are doing (and some data to back it up) to move on to the even-more-complicated step of going from prototype to production. At Netdata, we are currently trying to get to step 4, where we can first prototype some solutions on real-world data and come up with ways to measure progress.
This is where you, a valued member of the community, come in! We are looking to build a small pilot program and offer early access to Netdata research contributors who would stream some agents to a master node in our research data lake. Your data will stream to its own master node in our GCP project and only be used internally by Netdata for product research and development purposes. You can delete it at any time.
We can then use this real-world data to run experiments and try different problem formulations and algorithms to detect any anomalies that might exist. This way we can “battle test” as much as possible any ML-driven anomaly detection features we look to launch in the future with the help of the wider Netdata community.
This is not just a request for data contributions. We would also love to build an active community of Netdata users with an interest in machine learning and ideas and thoughts about how we can leverage it in various ways to redefine the future of monitoring! If you are interested in getting involved or finding out more, please contact us at [email protected].