Temporal Graph State Management System (TGSMS)

[csmangum]

• Temporal: Emphasizes the time-based aspect of the system.
• Graph: Highlights the graph-based structure for relationships.
• State Management: Indicates the core functionality of tracking and managing states over time.
• System: Signifies the comprehensive nature of the design, integrating various components.

[csmangum]

First step is to prototype the functionality that I want:

• Create a graph of the current period collection of agents (Arena)
• Easily extract the state as a graph for the ArenaState
• Connect nodes between the ArenaState with threads, which are temporal edges
• The state is a snapshot of the Arena at time of observation
• The TemporalState is the connected ArenaStates with the Arena itself the current and terminal graph, with previous states in order before it
• Thread types: Self (connects specific agents over ArenaStates), Genetic (parent and child)
• Easily create Threads when a state observation is taken from the Arena graph
• Easily trace backwards for any agent node in the Arena, through self or genetic threads into past ArenaStates
• Easily select ArenaStates in the past and traverse threads between any selected agent node. Forward and backward
• Easily traverse through ArenaStates, forward and backward

[csmangum]

When each arena state represents new agents, the Threads will be the parents of the new agents.

If an agent can persist more than one step, then a self thread connects it through its continuous ArenaStates

[csmangum]

Add encoding property to an ArenaState that is a lower dimensional representation of the graph state.

Use PCA at first with eventual Auto-Encoder to PCA for better encoding

[csmangum]

Have methods for Macro and Micro states

[csmangum]

Time is represented as a decimal cycle.step.

So cycle 12 and step 3, the time state should be 12.3
125.52 would be cycle 125 step 52
So ArenaStates can be clustered by cycle and be easily searched

Maybe floating point number to better handle very high values???