Revise Memory,Event,Percept Representations and Type Wrapping
toaomalkster opened this issue · 0 comments
toaomalkster commented
As per "Memory, Events and Percepts - A discussion on Representation" page.
TODO - phase 1:
- LongTermMemory story Events and Percepts as first-class objects.
- PerceptEvents are only used for historical memories of inputs that have been "perceived".
- LongTermMemory .search() returns List (where Object is always either Event or Percept)
- LongTermMemorySearchProcessor no longer needs to un-wrap PerceptEvents into Percepts.
- In first cut, MemoryEvent kept but changed to be able to store arbitrary objects, and then used to store List.
- FindMatchingConceptProcessor updated as needed when handling the MemoryEvent.
- Processors to return Events without timestamp, and with triggering Event(s) in references.
- Processors emit with their own totally ego-centric strength.
- AA uses references to decide on relative strength of the event.
- AA sets timestamp.
- Get rid of use of 'HANDLED' flag
- Should be able to create a very simple DSL for emitting Event that automatically tracks whether the event has already been emitted, taking that complexity out of the processors. ie: make processors more stateless and to just work immediately off what's in WM. I wonder how far this could be taken? See wiki page "Processors-like-Cloud-Lambda-Functions".
- Consider whether to remove the existence of MemoryEvent altogether
- Requires LTM results to be stored as individual items (un-collected) into WM.
TODO - phase 2:
TODO - phase 3: