making complexity simple
differentiable learning over millions of autonomous agents

Large Population Models (LPMs) help simulate million-size populations by designing realistic environments and capturing expressive indvidual behavior. Our goal is to "re-invent the census": built entirely in simulation, captured passively and used to protect country-scale populations. Our research is early but actively making an impact - winning awards at AI conferences and being deployed across the world. Learn more about LPMs here.

AgentTorch LPMs have four design principles:

• Scalability: AgentTorch models can simulate country-size populations in seconds on commodity hardware.
• Differentiability: AgentTorch models can differentiate through simulations with stochastic dynamics and conditional interventions, enabling gradient-based learning.
• Composition: AgentTorch models can compose with deep neural networks (eg: LLMs), mechanistic simulators (eg: mitsuba) or other LPMs. This helps describe agent behavior using LLMs, calibrate simulation parameters and specify expressive interaction rules.
• Generalization: AgentTorch helps simulate diverse ecosystems - humans in geospatial worlds, cells in anatomical worlds, autonomous avatars in digital worlds.

AgentTorch is building the future of decision engines - inside the body, around us and beyond!

The easiest way to install AgentTorch (v0.4.0) is from pypi:

> pip install agent-torch

AgentTorch is meant to be used in a Python 3.9 environment. If you have not installed Python 3.9, please do so first from python.org/downloads.

Install the most recent version from source using pip:

> pip install git+https://github.com/agenttorch/agenttorch

Some models require extra dependencies that have to be installed separately. For more information regarding this, as well as the hardware the project has been run on, please see docs/install.md.

The following section depicts the usage of existing models and population data to run simulations on your machine. It also acts as a showcase of the Agent Torch API.

A Jupyter Notebook containing the below examples can be found here.

# re-use existing models and population data easily
from agent_torch.models import covid
from agent_torch.populations import astoria

# use the executor to plug-n-play
from agent_torch.core.executor import Executor
from agent_torch.core.dataloader import LoadPopulation

# agent_"torch" works seamlessly with the pytorch API
from torch.optim import SGD

loader = LoadPopulation(astoria)
simulation = Executor(model=covid, pop_loader=loader)

simulation.init(SGD)
simulation.execute()

A detailed explanation of the architecture of the Agent Torch framework can be found here.

A tutorial on how to create a simple predator-prey model can be found in the tutorials/ folder.

from agent_torch.core.llm.archetype import Archetype
from agent_torch.core.llm.behavior import Behavior
from agent_torch.core.llm.backend import LangchainLLM
from agent_torch.populations import NYC

user_prompt_template = "Your age is {age} {gender},{unemployment_rate} the number of COVID cases is {covid_cases}."

# Using Langchain to build LLM Agents
agent_profile = "You are a person living in NYC. Given some info about you and your surroundings, decide your willingness to work. Give answer as a single number between 0 and 1, only."
llm_langchian = LangchainLLM(
    openai_api_key=OPENAI_API_KEY, agent_profile=agent_profile, model="gpt-3.5-turbo"
)

# Create an object of the Archetype class
# n_arch is the number of archetypes to be created. This is used to calculate a distribution from which the outputs are then sampled.
archetype = Archetype(n_arch=7)

# Create an object of the Behavior class
# You have options to pass any of the above created llm objects to the behavior class
# Specify the region for which the behavior is to be generated. This should be the name of any of the regions available in the populations folder.
earning_behavior = Behavior(
    archetype=archetype.llm(llm=llm_langchian, user_prompt=user_prompt_template), region=NYC
)

kwargs = {
    "month": "January",
    "year": "2020",
    "covid_cases": 1200,
    "device": "cpu",
    "current_memory_dir": "/path-to-save-memory",
    "unemployment_rate": 0.05,
}

output = earning_behavior.sample(kwargs)

Thank you for your interest in contributing! You can contribute by reporting and fixing bugs in the framework or models, working on new features for the framework, creating new models, or by writing documentation for the project.

Take a look at the contributing guide for instructions on how to setup your environment, make changes to the codebase, and contribute them back to the project.

AgentTorch models are being deployed across the globe.