Predicting or forecasting an event means the ability to determine in advance the time of its occurrence with a certain degree of confidence and precision. As defined by Morman et al the term anticipation implies more of an uncertainty as to when exactly an event will occur. This latter concept better fits the design of Falkonry’s recognition pipelines, which usually anticipate a behavior to occur within a certain time period after a condition is assessed without knowing its exact onset time.
Falkonry recognizes both frequency and spatial patterns present in time series. In other words, Falkonry takes into account how fast data varies and how much it varies over time. Moreover, it locates patterns across multiple variables, even though they may not be synchronously sampled.
No historical data is required. Falkonry can discover patterns in data as it is received and grow its pattern catalog over time. At the same time, it can apply the knowledge of such patterns to discover any new behavior that sets in as soon as that behavior occurs. Additionally, any critical behavior that occurs during the early stages of Falkonry deployment can be tagged with a condition name so that such a name is reported at the early sign of occurrence of that pattern.
No example facts are required. Falkonry can auto-discover patterns present in the data through a process called unsupervised learning. Such patterns are representatives of mathematical and temporal correlations. However, when a subject matter expert tags such patterns by providing facts, Falkonry defines a mapping between the patterns and domain behaviors. Only a handful of such examples are necessary to create a strong basis for Falkonry to later recognize those behaviors.
The best way to teach Falkonry about “normal” is by providing example facts of normal behavior. An example fact specifies the duration of time when a certain behavior was known to have present. Falkonry will then learn automatically about the ranges of values that are considered normal.
Falkonry analyzes the signs of distress from each and every time series. There is no requirement for time series signals to act in concert. At the same time, Falkonry will look at the relationships among time series in a multivariate sense, and that includes lead-lag relations.
A subject matter expert can learn the functions of Falkonry within an hour. The same expert can experiment with Falkonry and through our guidance can become proficient at defining and conducting experiments within 2-3 days.
Falkonry can analyze the behavior of individual entities in the context of a group. At the same time, Falkonry can discover patterns of behavior across an entire group where the behavior of each entity is considered as a signal time series. In the former case, Falkonry develops an entity behavior model whereas in the latter, a group behavior model is developed.
The Falkonry installation file is approximately 2 GB in size. It would take approximately 15 minutes to download and install this file on most business networks and systems.
You should have Docker version 1.12 on the system where you want to use Falkonry. Our installer is tested to work against Fedora, Redhat, Centos, and Ubuntu.
We endeavor to minimize the amount of consulting required to use Falkonry. That said, most organizations need a jump start to get going with Falkonry. With that in mind, Falkonry provides training – both in person and Web-based. Falkonry also provides hourly consulting support to help its customers map their data and requirements to pattern recognition.
One of the best ways to engage Falkonry is on a short-term pilot, during which we help your team both understand and apply Falkonry to your problem based on an objective and timeline.
Have a different question?
Here are two ways to contact us
Ask Us in Intercom
You can also interact with us online via Falkonry Software with the in-app chat. We are available around the clock to help with simple questions.