3.6.1. The Adaptive Model
MLPro provides the central template class Model for adaptive models. This bundles all properties important for machine learning on an abstract level. It represents the basis for all higher adaptive classes of the entire MLPro ecosystem and inherits its essential properties and possibilities for application-specific adjustments to them.
Performant Execution
As shown in the simplified class diagram above, the Model class is made up of numerous base classes of the lower levels through inheritance. So, from Layer 1 - Computation, it inherits the executability and asynchronous processing capabilities of class Task. In this way, it can also be combined in workflows to (parallelly/asynchronously) executable groups of models. From class EventHandler of the same level, it inherits the ability to raise events and forward them to registered event handlers.
Persistence
In particular, from Layer 0 - Elementary Functions, it inherits the ability of the LoadSave class to be able to be saved and reloaded. Other elementary capabilities such as logging, visualization, buffering of sample data, and referencing a scientific source are also fed in from this lower level.
Adaptivity
The Model class itself adds the ability to adapt. To this end, two mechanisms are introduced that support explicit adaptation based on external data and event-oriented adaptation. In both cases, the event C_EVENT_ADAPTED is raised, which can also be optionally handled as part of event handling. In this way, adaptation cascades can be triggered in a group of cooperating models.
Hyperparameters
Also, at the top layer 4 for machine learning, a system for hyperparameters is introduced and added to the Model class. These, in turn, take up the concepts of Dimension, Set/Space, and Element from Layer 2 - Mathematics.
Cross Reference