科研 Skill

Mesa3主体建模引擎

alterlab-abm-mesa

Builds agent-based models with the Mesa 3 AgentSet API, cell spaces, and batch_run sweeps for segregation, diffusion, and cooperation. Use for a Mesa 3.x ABM build or debug.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Agent-Based Modeling with Mesa 3 — Micro Rules, Macro Emergence

Skill type: ANALYSIS MODULE. Builds ABMs where macro patterns emerge from local agent rules (Schelling segregation, opinion dynamics, diffusion, cooperation). The discipline: use the current Mesa 3 API, and treat the model as a generative explanation that must be validated (swept, replicated, compared to a target pattern), not a toy that merely runs.

Core Mission

THE MODEL IS A THEORY OF HOW MACRO EMERGES FROM MICRO. USE MESA 3 CORRECTLY, THEN VALIDATE.

When to Use This Skill

  • "Build an agent-based model / Mesa model of [segregation, diffusion, opinion, cooperation]."
  • "Simulate many interacting heterogeneous agents and watch what emerges."
  • "Do a parameter sweep over my ABM and collect outcomes."

Does NOT Trigger

The request is really about…Route toWhy not this skill
Discrete-event / queueing / process simulationalterlab-simpyEvent-driven processes, not interacting agents on a grid/network.
Reinforcement learning (an agent learning a policy)alterlab-stable-baselines3Policy optimization, not generative social simulation.
Whether ABM is the right method at allalterlab-ssci-design-gateDesign routing, upstream.
System-dynamics / ODE compartment modelsalterlab-statistical-analysisAggregate dynamics, not agent-level.

Hand-off rule. When the request matches a row above, open your answer by naming that skill by its exact id (for example alterlab-simpy) and saying in one line why it fits better, then stop applying this skill's procedure. Name it even when you also need to ask for missing input. If only part of the request is out of scope, name the skill for that part and do only the in-scope part. A brief general pointer is fine; the named skill does the work.

Running the code. In Claude Code, run analysis code in a throwaway environment instead of installing packages: uv run --python ">=3.12" --with "mesa[network]>=3.4" python model.py (Mesa 3.4+ needs Python 3.12; import mesa needs networkx). This needs no install approval and leaves the user's environment untouched. If uv or the network is unavailable, say so and hand over the script; never report results you did not compute.

The Mesa 3 API (verified — do not ship Mesa 2 patterns)

Mesa 3 removed mesa.time schedulers (RandomActivation, etc.). Agents auto-register into model.agents; unique_id is auto-assigned; the model must call super().__init__(rng=...). Mesa 3.5 deprecated the seed= keyword in favor of rng= (same behavior, but seed= now emits a FutureWarning), so new code passes rng=.

import mesa

class MoneyAgent(mesa.Agent):
    def __init__(self, model):                 # NO unique_id argument in Mesa 3
        super().__init__(model)
        self.wealth = 1

    def step(self):
        if self.wealth > 0:
            other = self.random.choice(self.model.agents)
            other.wealth += 1
            self.wealth -= 1

class MoneyModel(mesa.Model):
    def __init__(self, n=10, rng=None):
        super().__init__(rng=rng)              # mandatory; rng= replaces seed= (Mesa 3.5)
        MoneyAgent.create_agents(self, n)      # bulk creation helper
        self.datacollector = mesa.DataCollector(
            model_reporters={"Gini": compute_gini},
            agent_reporters={"Wealth": "wealth"})

    def step(self):
        self.datacollector.collect(self)
        self.agents.shuffle_do("step")         # replaces RandomActivation
  • Scheduling: model.agents.shuffle_do("step") (random activation), model.agents.do("step") (fixed order), do("step") then do("advance") (simultaneous), model.agents_by_type[Type].shuffle_do("step") (staged). model.steps auto-increments.
  • Space: classic mesa.space.MultiGrid/SingleGrid/NetworkGrid (maintenance mode) with place_agent/move_agent; new cell space mesa.discrete_space.OrthogonalMooreGrid((w,h), torus=...) / OrthogonalVonNeumannGrid / HexGrid.
  • Batch sweeps: mesa.batch_run(MoneyModel, parameters={"n":[10,50,100]}, rng=range(20), max_steps=100, data_collection_period=-1) → list of dicts; each rng value seeds one replicate (iterations= is deprecated since Mesa 3.4).
  • Visualization: from mesa.visualization import SolaraViz, make_space_component, make_plot_component.

Full validated skeleton (Schelling + a network model), the space-API choice, and the validation checklist: references/mesa_patterns.md.

The validation discipline

An ABM that runs is not evidence. Report:

  1. Replication — multiple runs with different seeds; report the distribution of the macro outcome, not one run (fix rng= for reproducibility, vary it for the distribution).
  2. Parameter sweep — batch_run across the key parameters; show how the emergent outcome depends on them (phase transitions, tipping points).
  3. Pattern-oriented validation — does the model reproduce the target stylized fact it was built to explain (e.g. Schelling's high segregation from mild preferences)?
  4. Sensitivity — which assumptions drive the result; state them.

Output Template

MODEL:       <agents, their state, the local step rule; space = grid/network/none>
SCHEDULE:    <shuffle_do / do / staged> (Mesa 3 AgentSet)
MACRO OUTCOME: <the emergent quantity the DataCollector tracks>
VALIDATION:  <n seeds -> distribution; batch_run sweep; target pattern reproduced?>
CLAIM SCOPE: generative (a sufficient micro mechanism), NOT the unique/true mechanism

References

  • references/mesa_patterns.md — validated Mesa 3 skeletons (grid + network), space-API selection, batch_run sweep, validation checklist.

Part of the AlterLab Academic Skills suite.

示例提示词

帮我用Mesa3的AgentSet API搭建一个社会隔离涌现的主体仿真模型。

所属插件

社会学、政治学与城市研究面向社会科学研究:问卷与访谈设计、质性编码、文本即数据、社会网络、ABM、比较政治与选举数据、RCT 与 DID、城市与区域分析。

安装该插件会一并获得同领域的 58 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

Mesa3主体建模引擎能做什么?

Builds agent-based models with the Mesa 3 AgentSet API, cell spaces, and batch_run sweeps for segregation, diffusion, and cooperation. Use for a Mesa 3.x ABM build or debug。

如何安装Mesa3主体建模引擎?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「Mesa3主体建模引擎」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

Mesa3主体建模引擎需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

Mesa3主体建模引擎经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。