gtlab 0.2.5

Teaching toolkit for the ELTE Game Theory course: solvers, game library, and Jupyter-friendly visualizations.

gtlab - Game Theory Lab

A teaching toolkit for the ELTE Game Theory course. It consolidates the solvers, example games, and Jupyter visualizations that previously lived (duplicated) in six separate Colab notebooks into one installable, extensible package.

Install

pip install -e ".[full]"     # full = nashpy + pulp extras (recommended in Colab)
pip install -e .             # core only (numpy / scipy / matplotlib / pandas)

Quick start

import gtlab
gtlab.apply_rc()                         # shared matplotlib styling (call once)

from gtlab.games import prisoners_dilemma
prisoners_dilemma().solve()              # annotated bimatrix in a Jupyter cell

Build your own game:

import numpy as np
from gtlab import NormalFormGame

g = NormalFormGame(
    A=np.array([[3, 0], [5, 1]]),
    B=np.array([[3, 5], [0, 1]]),
    row_actions=["Cooperate", "Defect"],
    col_actions=["Cooperate", "Defect"],
    name="My Game",
)
g.explain()

Architecture

The design separates the three concerns that were tangled together in the notebooks:

gtlab/
├── core/      game classes - hold data, expose a thin API
│   ├── normal_form.py      NormalFormGame
│   ├── zero_sum.py         ZeroSumGame
│   ├── correlated.py       CorrelatedGame
│   ├── stochastic.py       StochasticGame
│   ├── extensive_form.py   ExtensiveFormGame
│   └── bayesian.py         PostedPrice, FirstPriceAuction, SecondPriceAuction
├── solvers/   pure algorithms - numpy in, numpy/dict out, no display
│   ├── best_response.py    nash.py          dominance.py
│   ├── pareto.py           linprog.py       value_iteration.py
│   ├── welfare.py          learning.py      correlated.py
├── viz/       display layer - ONE theme, formatters, HTML, plots
│   ├── theme.py  format.py  html.py  plots.py
└── games/     ready-made example games + REGISTRY

Rule of thumb: math goes in solvers/, rendering goes in viz/, and the classes in core/ just wire them together. A theme tweak is one edit in viz/theme.py; a new algorithm is one file in solvers/; a new example is one factory in games/.

Extending

To add…	…do this
a new example game	write a factory in games/, add it to REGISTRY
a new algorithm	add a pure function in solvers/, export it
a new game type	add a class in core/ that calls solvers + viz
restyle output	edit viz/theme.py (colors / CSS / rcParams)

Tests

pip install -e ".[dev]"
pytest

The solver layer has golden-value tests so refactors stay behavior-preserving.

Notebook migration

Each lecture notebook drops its ~2,000-line engine class and imports from gtlab instead, keeping only narrative and example-specific calls. As of 0.2.0 the package covers the full notebook surface: every solver, plot (plot_tree, plot_mixed, plot_frontier, ...), parameter sweep, styled walkthrough (explain, iesds_explain, *_detail, verify_*), Monte-Carlo simulate, and the GeneralSumSG / CheapTalkGame models. The methods the notebooks relied on (.summary(), .solve(), .explain(), .plot_*()) are preserved on the core classes.

# old: 1,800 lines of NormalFormGame defined inline
# new:
from gtlab.games import prisoners_dilemma
prisoners_dilemma().solve(show_br=True, show_ne=True)