- WTF Core +
Need a classification framework which allows several trained/traditional models to work in unison and "vote" for what they believe to be the correct classification. The ultimate goal for this is to provide a single input, which is passed down to many members who vote on what the classification based on data fed to them, then bubbled up back to the caller.
High level controller of multiple models, and when provided data, will pass along to all models in the group. When a manager is asked to give a report on the current state of data, he will ask all models for their input on the classification, but will not treat all models as equal (weighted trustworthiness). Based on their weights, will give the caller the best guess of what the classification is, as well as a guid which can be used to provide back to the manager stating what the correct classification actually was. This feedback process will allow the manager to re-adjust biases from his collection of models and hopefully adjust to the conditions of the environment, without having to re-train models.
The common elements that a manager must have:
-What data the manager operates with -What are the classifications that the manager can provide -Feed Data -Classify Method -Provide Feedback (guid, and what the actual classification was supposed to be)
A model can use any sort of mechanism to determine what sort of classification the current state the data provided to it is in.
The common elements that a model must have:
-What data the model operates with -What are the classifications that the model can provide -Feed Data -Classify Method