ABMForge Documentation

ABMForge is a lightweight, reproducible, experiment-native agent-based modeling framework for Python.

It is designed for researchers, educators, model developers, data scientists, and students who want to build agent-based simulations that are reproducible, analyzable, and extensible.

Core Principles

ABMForge is built around four principles:

Reproducibility

Simulation runs should be repeatable through explicit seeds, run metadata, manifests, and snapshots.

Experiment-Native Design

Experiments are first-class concepts rather than afterthoughts.

Dataset-First Outputs

Simulation outputs should be easy to analyze using standard data tools.

Lightweight Python Architecture

The core framework should remain easy to understand, easy to test, and easy to extend.

Main Components

Core Modeling

Agent
Model
AgentCollection

Spaces

GridWorld
NetworkSpace
ContinuousSpace
GISSpace

Scheduling

SequentialActivation
RandomActivation
SimultaneousActivation
StagedActivation

Experiments

Scenario
Experiment
ParameterGrid
ExperimentResult

Data and Reproducibility

Recorder
Dataset
CSV export
JSON/JSONL export
Reproducibility manifest
Snapshot read/write helpers

Analysis

SensitivityAnalysis
Optional SALib integration
Sobol sampling
Morris sampling

Visualization

plot_timeseries
plot_multiple_runs
plot_grid

Example Gallery

ABMForge currently includes:

Wealth model
Schelling segregation
SIR epidemic
Sugarscape
Parameter sweep
GIS example