Operating Concepts
Creating a model in Cogynt Authoring usually follows a sequential, top-down pattern where users begin with the highest-level view and work down to increasing levels of specificity as needed.
This section outlines the main steps of the modeling process, describing common tasks during each stage. For more information about each stage, see the corresponding sections of Developing Models.
Creating Projects
Every model starts with the creation of a project, which encapsulates all of a model's data. For more information, see Creating and Opening Projects.
A model in one project cannot reference artifacts in another project. However, Cogynt Authoring does allow exporting selected artifacts in one project and importing them into another project. The imported artifacts are duplicated and given new identifiers to prevent conflicts with the originals. For more information, see Exporting and Importing Authoring Files.
For more information about model artifacts, including projects, see Model Artifacts.
System Authoring
Users engage with the system view first. In this view:
- Empty event patterns are created.
- Data flow from one event pattern to another is established by drawing a link between them.
- Cogynt Authoring prompts the user to pick what kind of data (called an “event type”) is valid for that data flow. If the desired event type does not exist, it can be created as needed. As part of this process, users must pick an existing user data schema (or create a new one) for the event type to use.
- Existing event patterns with predefined elements or outcomes have the option of choosing an event type referenced by those elements or outcomes from either the source or destination event pattern.
- Once an event type is selected, Cogynt Authoring creates any missing model artifacts (an element for input or an outcome for output) in both event patterns to establish a valid data flow connection between the two.
For more information about working with the system view, see System Authoring.
Event types and user data schemas can be created through their respective management utilities at any time outside of all views.
Event Pattern Authoring
Work with pattern detection logic occurs in the event pattern view.
To reach the event pattern view:
- Select an event pattern in the system view.
- Double-click the event pattern to drill down.
In the event pattern view, the following tasks are performed:
- Elements and outcomes needed for pattern detection logic are created using the tool palette.
- Constraints are created by drawing links between elements.
- From a constraint:
- Conditions for fields between the two connected elements can be specified for Cogynt to evaluate.
- Constraint computations for value comparisons between the connected elements can be provided.
- For an outcome:
- A risk distribution can be optionally created.
- Its windowing feature can be enabled.
For more information about working with the event pattern view, see Event Pattern Authoring.
Outcome Computations Authoring
Since outcomes need to generate data payloads for output, logic to compute their field values must be provided. This is accomplished in the outcome computations view by creating data mappings from the fields of input event types associated with elements to the fields of output event types associated with the outcome.
To reach the outcome computations view:
- Double-click (drill down through) an outcome in the event pattern view.
- The outcome computations view opens.
In the outcome computations view, data mappings between fields of one or more input event types to fields of output event types are specified:
- For the simplest cases, a link can be drawn from an input field to a compatible output field to indicate a field value transfer.
- For cases requiring more complex computations for an output field based on the values of input fields, Cogynt Authoring provides various computation elements (such as constants, functions, arithmetic operators, and so on).
For information about working with the outcome computations view, see Outcome Computations Authoring.
The constraint computations view is similar to the outcome computations view. The key differences are:
- The constraint computations view can only generate one single output value for the constraint end where it is defined.
- The outcome computations view generates a full payload for output event types.
The constraint computations view can only be accessed by drilling down through a constraint. For more information about the constraint computations view, see Constraint Computations Authoring.
Lexicon Authoring
If a problem domain requires some form of text filtering against the input data stream, a lexicon (a dictionary of terms organized in a hierarchical structure) can be created for this purpose using the Lexicon Management utility.
To access the Lexicon Management utility:
- Click the system panel (☰) icon to open the system panel.
- Under Other Tasks, click Lexicons.
Once a lexicon is created, you can define a lexicon filter for a field in an event type.
For more information about working with lexicons, see Lexicon Authoring.
Deploying and Monitoring Models
For the final stage of model development, the model is deployed to an Apache Flink cluster powered by the Cogynt Hierarchical Complex Event Processing (HCEP) engine:
- A deployment target must be created that specifies:
- The connection information for the model’s data sources.
- Any custom Flink configurations.
- Any logging configurations.
- When a deployment is created, a deployment target for it to use must also be selected.
A deployment is the model artifact that indicates the event patterns for Cogynt to deploy.
For more information about model artifacts, see Model Artifacts.
Once the model (or some parts of the model) is deployed, its status and states can be monitored through the Deployment Management utility in conjunction with some other tools such as Cogynt Workstation, the Flink console, Imply visualization tools, and the like. These are bundled as part of Cogility's Cogynt solution.
For more information about deploying models, see Deploying Models.
For more information about monitoring deployments, see Monitoring Deployments.