clarkkitchen22's picture
Initial GeoBot Forecasting Framework commit
484e3bc
"""
Agent-Based Modeling for Geopolitical Simulation
Models individual actors (states, organizations, leaders) and their interactions.
"""
import numpy as np
from typing import Dict, List, Optional, Any, Tuple
from dataclasses import dataclass, field
from enum import Enum
class AgentType(Enum):
"""Types of geopolitical agents."""
STATE = "state"
ORGANIZATION = "organization"
LEADER = "leader"
ALLIANCE = "alliance"
@dataclass
class AgentState:
"""State variables for an agent."""
position: np.ndarray # Position in feature space
resources: float = 1.0
power: float = 1.0
hostility: float = 0.0
cooperation: float = 0.5
stability: float = 1.0
custom: Dict[str, float] = field(default_factory=dict)
class GeopoliticalAgent:
"""
Represents a geopolitical actor.
Agents have internal state, decision-making logic, and
interact with other agents and the environment.
"""
def __init__(
self,
agent_id: str,
agent_type: AgentType,
initial_state: AgentState
):
"""
Initialize agent.
Parameters
----------
agent_id : str
Unique agent identifier
agent_type : AgentType
Type of agent
initial_state : AgentState
Initial state
"""
self.agent_id = agent_id
self.agent_type = agent_type
self.state = initial_state
self.history: List[AgentState] = [initial_state]
self.relationships: Dict[str, float] = {} # {agent_id: relationship_strength}
def update_state(self, **kwargs) -> None:
"""Update agent state variables."""
for key, value in kwargs.items():
if hasattr(self.state, key):
setattr(self.state, key, value)
else:
self.state.custom[key] = value
def decide_action(
self,
environment: Dict[str, Any],
other_agents: List['GeopoliticalAgent']
) -> Dict[str, Any]:
"""
Decide on action based on current state and environment.
Parameters
----------
environment : dict
Environmental factors
other_agents : list
Other agents in the system
Returns
-------
dict
Chosen action
"""
# Simple decision logic (can be made more sophisticated)
action = {
'type': 'none',
'target': None,
'intensity': 0.0
}
# Check for threats
threats = [
agent for agent in other_agents
if self.relationships.get(agent.agent_id, 0) < -0.5
and agent.state.power > self.state.power * 0.8
]
if threats and self.state.hostility > 0.5:
# Consider conflict
action = {
'type': 'escalate',
'target': threats[0].agent_id,
'intensity': self.state.hostility
}
elif self.state.cooperation > 0.7:
# Seek cooperation
potential_partners = [
agent for agent in other_agents
if self.relationships.get(agent.agent_id, 0) > 0.3
]
if potential_partners:
action = {
'type': 'cooperate',
'target': potential_partners[0].agent_id,
'intensity': self.state.cooperation
}
return action
def interact(self, other_agent: 'GeopoliticalAgent', action: Dict[str, Any]) -> None:
"""
Interact with another agent.
Parameters
----------
other_agent : GeopoliticalAgent
Other agent
action : dict
Action to perform
"""
action_type = action['type']
intensity = action['intensity']
if action_type == 'cooperate':
# Strengthen relationship
current_rel = self.relationships.get(other_agent.agent_id, 0)
self.relationships[other_agent.agent_id] = min(1.0, current_rel + 0.1 * intensity)
# Mutual benefit
self.state.resources += 0.05 * intensity
other_agent.state.resources += 0.05 * intensity
elif action_type == 'escalate':
# Weaken relationship
current_rel = self.relationships.get(other_agent.agent_id, 0)
self.relationships[other_agent.agent_id] = max(-1.0, current_rel - 0.2 * intensity)
# Conflict effects
power_ratio = self.state.power / (other_agent.state.power + 1e-6)
if power_ratio > 1:
self.state.resources += 0.1 * intensity
other_agent.state.resources -= 0.15 * intensity
other_agent.state.stability -= 0.1 * intensity
else:
self.state.resources -= 0.1 * intensity
self.state.stability -= 0.05 * intensity
def save_state(self) -> None:
"""Save current state to history."""
self.history.append(AgentState(
position=self.state.position.copy(),
resources=self.state.resources,
power=self.state.power,
hostility=self.state.hostility,
cooperation=self.state.cooperation,
stability=self.state.stability,
custom=self.state.custom.copy()
))
class AgentBasedModel:
"""
Agent-based model for geopolitical simulation.
Manages multiple agents and their interactions over time.
"""
def __init__(self):
"""Initialize agent-based model."""
self.agents: Dict[str, GeopoliticalAgent] = {}
self.environment: Dict[str, Any] = {}
self.time: int = 0
def add_agent(self, agent: GeopoliticalAgent) -> None:
"""
Add agent to model.
Parameters
----------
agent : GeopoliticalAgent
Agent to add
"""
self.agents[agent.agent_id] = agent
def remove_agent(self, agent_id: str) -> None:
"""
Remove agent from model.
Parameters
----------
agent_id : str
Agent ID to remove
"""
if agent_id in self.agents:
del self.agents[agent_id]
def set_environment(self, **kwargs) -> None:
"""Set environmental variables."""
self.environment.update(kwargs)
def step(self) -> None:
"""
Execute one time step of simulation.
All agents make decisions and interact.
"""
self.time += 1
# Phase 1: All agents decide actions
actions = {}
other_agents_list = list(self.agents.values())
for agent in self.agents.values():
action = agent.decide_action(self.environment, other_agents_list)
actions[agent.agent_id] = action
# Phase 2: Execute actions
for agent_id, action in actions.items():
agent = self.agents[agent_id]
if action['type'] != 'none' and action['target'] is not None:
if action['target'] in self.agents:
target = self.agents[action['target']]
agent.interact(target, action)
# Phase 3: Environmental updates
for agent in self.agents.values():
# Resource growth
agent.state.resources *= (1 + 0.01 * agent.state.stability)
# Power calculation
agent.state.power = agent.state.resources * agent.state.stability
# Add noise
agent.state.hostility += np.random.normal(0, 0.05)
agent.state.hostility = np.clip(agent.state.hostility, 0, 1)
agent.state.cooperation += np.random.normal(0, 0.05)
agent.state.cooperation = np.clip(agent.state.cooperation, 0, 1)
# Save state
agent.save_state()
def run(self, n_steps: int) -> None:
"""
Run simulation for multiple steps.
Parameters
----------
n_steps : int
Number of time steps
"""
for _ in range(n_steps):
self.step()
def get_agent_trajectories(self, agent_id: str) -> Dict[str, List[float]]:
"""
Get historical trajectories for an agent.
Parameters
----------
agent_id : str
Agent ID
Returns
-------
dict
Trajectories of state variables
"""
if agent_id not in self.agents:
return {}
agent = self.agents[agent_id]
history = agent.history
return {
'resources': [s.resources for s in history],
'power': [s.power for s in history],
'hostility': [s.hostility for s in history],
'cooperation': [s.cooperation for s in history],
'stability': [s.stability for s in history]
}
def get_system_state(self) -> Dict[str, Any]:
"""
Get current state of entire system.
Returns
-------
dict
System state
"""
return {
'time': self.time,
'n_agents': len(self.agents),
'total_resources': sum(a.state.resources for a in self.agents.values()),
'mean_hostility': np.mean([a.state.hostility for a in self.agents.values()]),
'mean_cooperation': np.mean([a.state.cooperation for a in self.agents.values()]),
'mean_stability': np.mean([a.state.stability for a in self.agents.values()])
}
def analyze_network(self) -> Dict[str, Any]:
"""
Analyze the network of relationships.
Returns
-------
dict
Network metrics
"""
import networkx as nx
# Build network
G = nx.Graph()
for agent in self.agents.values():
G.add_node(agent.agent_id)
for agent in self.agents.values():
for other_id, strength in agent.relationships.items():
if strength > 0.1: # Only positive relationships
G.add_edge(agent.agent_id, other_id, weight=strength)
# Compute metrics
return {
'n_nodes': G.number_of_nodes(),
'n_edges': G.number_of_edges(),
'density': nx.density(G),
'average_clustering': nx.average_clustering(G) if G.number_of_edges() > 0 else 0,
'connected_components': nx.number_connected_components(G)
}