Meta agents

mesa/examples/__init__.py

@@ -2,6 +2,7 @@
 from mesa.examples.advanced.pd_grid.model import PdGrid
 from mesa.examples.advanced.sugarscape_g1mt.model import SugarscapeG1mt
 from mesa.examples.advanced.wolf_sheep.model import WolfSheep
+from mesa.examples.basic.alliance_formation_model.model import MultiLevelAllianceModel
 from mesa.examples.basic.boid_flockers.model import BoidFlockers
 from mesa.examples.basic.boltzmann_wealth_model.model import BoltzmannWealth
 from mesa.examples.basic.conways_game_of_life.model import ConwaysGameOfLife
@@ -13,6 +14,7 @@
     "BoltzmannWealth",
     "ConwaysGameOfLife",
     "EpsteinCivilViolence",
+    "MultiLevelAllianceModel",
     "PdGrid",
     "Schelling",
     "SugarscapeG1mt",

mesa/examples/basic/alliance_formation_model/Readme.md

@@ -0,0 +1,40 @@
+# Alliance Formation Model
+
+## Summary
+This model demonstrates Mesa's ability to dynamically create new classes of agents that are composed of existing agents. These meta-agents  inherits functions and attributes from their sub-agents and users can specify new functionality or attributes they want the meta agent to have. For example, if a user is doing a factory simulation with autonomous systems, each major component of that system can be a sub-agent of the overall robot agent. Or, if someone is doing a simulation of an organization, individuals can be part of different organizational units that are working for some purpose.
+
+To provide a simple demonstration of this capability is an alliance formation model.
+
+In this simulation n agents are created, who have two attributes (1) power and (2) preference. Each attribute is a number between 0 and 1 over a gaussian distribution. Agents then randomly select other agents and use the [bilateral shapley value](https://en.wikipedia.org/wiki/Shapley_value) to determine if they should form an alliance. If the expected utility support an alliances, the agent creates a meta-agent. Subsequent steps may add agents to the meta-agent, create new instances of similar hierarchy, or create a new hierarchy level where meta-agents form an alliance of meta-agents. In this visualization of this model a new meta-agent hierarchy will be a larger node and a new color.
+
+In its current configuration, agents being part of multiple meta-agents is not supported
+
+## Installation
+
+This model requires Mesa's recommended install and scipy
+```
+    $ pip install mesa[rec]
+```
+
+## How to Run
+
+To run the model interactively, in this directory, run the following command
+
+```
+    $ solara run app.py
+```
+
+## Files
+
+* ``model.py``: Contains creation of agents, the network and management of agent execution.
+* ``agents.py``: Contains logic for forming alliances and creation of new agents
+* ``app.py``: Contains the code for the interactive Solara visualization.
+
+## Further Reading
+
+The full tutorial describing how the model is built can be found at:
+https://mesa.readthedocs.io/en/latest/tutorials/intro_tutorial.html
+
+An example of the bilateral shapley value in another model:
+[Techno-Social Energy Infrastructure Siting: Sustainable Energy Modeling Programming (SEMPro)](https://www.jasss.org/16/3/6.html)
+

mesa/examples/basic/alliance_formation_model/agents.py

@@ -0,0 +1,20 @@
+import mesa
+
+
+class AllianceAgent(mesa.Agent):
+    """
+    Agent has three attributes power (float), position (float) and level (int)
+
+    """
+
+    def __init__(self, model, power, position, level=0):
+        super().__init__(model)
+        self.power = power
+        self.position = position
+        self.level = level
+
+    """
+    For this demo model agent only need attributes.
+
+    More complex models could have functions that define agent behavior.
+    """

mesa/examples/basic/alliance_formation_model/app.py

@@ -0,0 +1,72 @@
+import matplotlib.pyplot as plt
+import networkx as nx
+import solara
+from matplotlib.figure import Figure
+
+from mesa.examples.basic.alliance_formation_model.model import MultiLevelAllianceModel
+from mesa.visualization import SolaraViz
+from mesa.visualization.utils import update_counter
+
+model_params = {
+    "seed": {
+        "type": "InputText",
+        "value": 42,
+        "label": "Random Seed",
+    },
+    "n": {
+        "type": "SliderInt",
+        "value": 50,
+        "label": "Number of agents:",
+        "min": 10,
+        "max": 100,
+        "step": 1,
+    },
+}
+
+# Create visualization elements. The visualization elements are solara components
+# that receive the model instance as a "prop" and display it in a certain way.
+# Under the hood these are just classes that receive the model instance.
+# You can also author your own visualization elements, which can also be functions
+# that receive the model instance and return a valid solara component.
+
+
+@solara.component
+def plot_network(model):
+    update_counter.get()
+    g = model.network
+    pos = nx.fruchterman_reingold_layout(g)
+    fig = Figure()
+    ax = fig.subplots()
+    labels = {agent.unique_id: agent.unique_id for agent in model.agents}
+    node_sizes = [g.nodes[node]["size"] for node in g.nodes]
+    node_colors = [g.nodes[node]["size"] for node in g.nodes()]
+
+    nx.draw(
+        g,
+        pos,
+        node_size=node_sizes,
+        node_color=node_colors,
+        cmap=plt.cm.coolwarm,
+        labels=labels,
+        ax=ax,
+    )
+
+    solara.FigureMatplotlib(fig)
+
+
+# Create initial model instance
+model = MultiLevelAllianceModel(50)
+
+# Create the SolaraViz page. This will automatically create a server and display the
+# visualization elements in a web browser.
+# Display it using the following command in the example directory:
+# solara run app.py
+# It will automatically update and display any changes made to this file
+
+page = SolaraViz(
+    model,
+    components=[plot_network],
+    model_params=model_params,
+    name="Alliance Formation Model",
+)
+page  # noqa

mesa/examples/basic/alliance_formation_model/model.py

@@ -0,0 +1,180 @@
+import networkx as nx
+import numpy as np
+
+import mesa
+from mesa.examples.basic.alliance_formation_model.agents import AllianceAgent
+from mesa.experimental.meta_agents.meta_agent import find_combinations
+from mesa.experimental.meta_agents.multi_levels import multi_level_agents
+
+
+class MultiLevelAllianceModel(mesa.Model):
+    """
+    Model for simulating multi-level alliances among agents.
+    """
+
+    def __init__(self, n=50, mean=0.5, std_dev=0.1, seed=42):
+        """
+        Initialize the model.
+
+        Args:
+            n (int): Number of agents.
+            mean (float): Mean value for normal distribution.
+            std_dev (float): Standard deviation for normal distribution.
+            seed (int): Random seed.
+        """
+        super().__init__(seed=seed)
+        self.population = n
+        self.network = nx.Graph()  # Initialize the network
+        self.datacollector = mesa.DataCollector(model_reporters={"Network": "network"})
+
+        # Create Agents
+        power = self.rng.normal(mean, std_dev, n)
+        power = np.clip(power, 0, 1)
+        position = self.rng.normal(mean, std_dev, n)
+        position = np.clip(position, 0, 1)
+        AllianceAgent.create_agents(self, n, power, position)
+        agent_ids = [
+            (agent.unique_id, {"size": 300, "level": 0}) for agent in self.agents
+        ]
+        self.network.add_nodes_from(agent_ids)
+
+    def add_link(self, meta_agent, agents):
+        """
+        Add links between a meta agent and its constituent agents in the network.
+
+        Args:
+            meta_agent (MetaAgent): The meta agent.
+            agents (list): List of agents.
+        """
+        for agent in agents:
+            self.network.add_edge(meta_agent.unique_id, agent.unique_id)
+
+    def calculate_shapley_value(self, agents):
+        """
+        Calculate the Shapley value of the two agents.
+
+        Args:
+            agents (list): List of agents.
+
+        Returns:
+            tuple: Potential utility, new position, and level.
+        """
+        positions = agents.get("position")
+        new_position = 1 - (max(positions) - min(positions))
+        potential_utility = agents.agg("power", sum) * 1.2 * new_position
+
+        value_0 = 0.5 * agents[0].power + 0.5 * (potential_utility - agents[1].power)
+        value_1 = 0.5 * agents[1].power + 0.5 * (potential_utility - agents[0].power)
+
+        if value_0 > agents[0].power and value_1 > agents[1].power:
+            if agents[0].level > agents[1].level:
+                level = agents[0].level
+            elif agents[0].level == agents[1].level:
+                level = agents[0].level + 1
+            else:
+                level = agents[1].level
+
+            return potential_utility, new_position, level
+
+    def only_best_combination(self, combinations):
+        """
+        Filter to keep only the best combination for each agent.
+
+        Args:
+            combinations (list): List of combinations.
+
+        Returns:
+            dict: Unique combinations.
+        """
+        best = {}
+        # Determine best option for EACH agent
+        for group, value in combinations:
+            agent_ids = sorted(group.get("unique_id"))  # by default is bilateral
+            # Deal with all possibilities
+            if (
+                agent_ids[0] not in best and agent_ids[1] not in best
+            ):  # if neither in add both
+                best[agent_ids[0]] = [group, value, agent_ids]
+                best[agent_ids[1]] = [group, value, agent_ids]
+            elif (
+                agent_ids[0] in best and agent_ids[1] in best
+            ):  # if both in, see if both would be trading up
+                if (
+                    value[0] > best[agent_ids[0]][1][0]
+                    and value[0] > best[agent_ids[1]][1][0]
+                ):
+                    # Remove the old alliances
+                    del best[best[agent_ids[0]][2][1]]
+                    del best[best[agent_ids[1]][2][0]]
+                    # Add the new alliance
+                    best[agent_ids[0]] = [group, value, agent_ids]
+                    best[agent_ids[1]] = [group, value, agent_ids]
+            elif (
+                agent_ids[0] in best
+            ):  # if only agent_ids[0] in, see if it would be trading up
+                if value[0] > best[agent_ids[0]][1][0]:
+                    # Remove the old alliance for agent_ids[0]
+                    del best[best[agent_ids[0]][2][1]]
+                    # Add the new alliance
+                    best[agent_ids[0]] = [group, value, agent_ids]
+                    best[agent_ids[1]] = [group, value, agent_ids]
+            elif (
+                agent_ids[1] in best
+            ):  # if only agent_ids[1] in, see if it would be trading up
+                if value[0] > best[agent_ids[1]][1][0]:
+                    # Remove the old alliance for agent_ids[1]
+                    del best[best[agent_ids[1]][2][0]]
+                    # Add the new alliance
+                    best[agent_ids[0]] = [group, value, agent_ids]
+                    best[agent_ids[1]] = [group, value, agent_ids]
+
+        # Create a unique dictionary of the best combinations
+        unique_combinations = {}
+        for group, value, agents_nums in best.values():
+            unique_combinations[tuple(agents_nums)] = [group, value]
+
+        return unique_combinations.values()
+
+    def step(self):
+        """
+        Execute one step of the model.
+        """
+        # Get all other agents of the same type
+        agent_types = list(self.agents_by_type.keys())
+
+        for agent_type in agent_types:
+            similar_agents = self.agents_by_type[agent_type]
+
+            # Find the best combinations using find_combinations
+            if (
+                len(similar_agents) > 1
+            ):  # only form alliances if there are more than 1 agent
+                combinations = find_combinations(
+                    self,
+                    similar_agents,
+                    size=2,
+                    evaluation_func=self.calculate_shapley_value,
+                    filter_func=self.only_best_combination,
+                )
+
+                for alliance, attributes in combinations:
+                    class_name = f"MetaAgentLevel{attributes[2]}"
+                    meta = multi_level_agents(
+                        self,
+                        class_name,
+                        alliance,
+                        meta_attributes={
+                            "level": attributes[2],
+                            "power": attributes[0],
+                            "position": attributes[1],
+                        },
+                    )
+
+                    # Update the network if a new meta agent instance created
+                    if meta:
+                        self.network.add_node(
+                            meta.unique_id,
+                            size=(meta.level + 1) * 300,
+                            level=meta.level,
+                        )
+                        self.add_link(meta, meta.subset)

mesa/experimental/__init__.py

@@ -15,6 +15,6 @@
     - Features graduate from experimental status once their APIs are stabilized
 """
 
-from mesa.experimental import cell_space, devs, mesa_signals
+from mesa.experimental import cell_space, devs, mesa_signals, meta_agents
 
-__all__ = ["cell_space", "devs", "mesa_signals"]
+__all__ = ["cell_space", "devs", "mesa_signals", "meta_agents"]

mesa/experimental/meta_agents/__init__.py

@@ -0,0 +1,26 @@
+"""This method is for dynamically creating new agents (meta-agents).
+
+Meta-agents are defined as agents composed of existing agents.
+
+Meta-agents are created dynamically with a pointer to the model, name of the meta-agent,,
+iterable of agents to belong to the new meta-agents, any new functions for the meta-agent,
+any new attributes for the meta-agent, whether to retain sub-agent functions,
+whether to retain sub-agent attributes.
+
+Examples of meta-agents:
+- An autonomous car where the subagents are the wheels, sensors,
+battery, computer etc. and the meta-agent is the car itself.
+- A company where the subagents are employees, departments, buildings, etc.
+- A city where the subagents are people, buildings, streets, etc.
+
+Currently meta-agents are restricted to one parent agent for each subagent/
+one meta-agent per subagent.
+
+Goal is to assess usage and expand functionality.
+
+"""
+
+from .meta_agent import MetaAgent
+from .multi_levels import multi_level_agents
+
+__all__ = ["MetaAgent", "multi_level_agents"]