What algorithms are included?

The framework supports A*, BFS, DFS, UCS, IDA*, Hill Climbing, Simulated Annealing, Genetic Algorithm, Backtracking with MRV/MAC, Minimax, AlphaBeta, MCTS, and IS-MCTS across search, optimization, constraint satisfaction, and game domains.

How do I run the demos?

Install dependencies (pygame, matplotlib, numpy), then run `python main.py` to launch an interactive CLI menu or `python -m demo. --algo --vis` to run individual demos with visualization.

Can I add my own algorithm or domain?

Yes; the framework is designed for contributions. You can inherit from base classes (SearchAlgorithm, GameState, etc.) in the respective modules, or add new CSP heuristics and custom Pygame visualizers following the standardized patterns.

AI Coding AssistantsOpen-Source AIHacker News3h agoAug 15, 2026, 10:00 JST4 min read

Python AI lab open-sources classical algorithms framework

Send on LINE

Key takeaway

A new open-source Python framework has released classical AI algorithms—search, optimization, constraint satisfaction, and game-playing methods—packaged with interactive visualizers and a CLI launcher for 18 demos.
Designed for learning and research, the modular architecture lets users swap algorithms and domains, run benchmarks, and contribute new solvers; it is available under the MIT License.

3 Key Points

What happened
A GitHub repository has released a modular Python framework implementing classical AI algorithms—including search (A*, BFS, DFS), optimization (genetic algorithms, simulated annealing), constraint satisfaction, and adversarial game-playing—with 18 interactive demos, visualizers, and a CLI launcher.
Why it matters
The framework targets visual learners, researchers, and developers seeking hands-on understanding of foundational AI paradigms. It bundles transparent visualizers, automated benchmarking, and pluggable solver engines in a single codebase, reducing friction for teaching and experimentation without requiring ML infrastructure.
What to watch
The project invites open-source contributions (new algorithms, heuristics, game domains, visualizers) and provides benchmarking pipelines that generate performance reports and charts; source code and installation instructions are available via pip and direct Python module execution.

Summaries like this, in your inbox every morning.

Get news like this every morning — free

In Depth

Read the full story

The AI-Lab is a from-scratch Python implementation of classical artificial intelligence algorithms organized around four pillars: graph search and pathfinding, local search and continuous optimization, constraint satisfaction problems (CSP), and adversarial game theory.

The search suite includes standard uninformed algorithms (BFS, DFS, UCS) and informed heuristic search (A*, IDA*), demonstrated across domains such as maze navigation (with Manhattan and Euclidean heuristics), the 8-puzzle sliding tile problem (using disjoint pattern databases), map routing (Romanian city example), Sokoban box-pushing, and online real-time learning (LRTA*). Optimization covers local search variants—Hill Climbing, Simulated Annealing, Local Beam Search, and Genetic Algorithms—applied to the Traveling Salesman Problem and N-Queens. CSP solvers implement backtracking with constraint propagation (MRV heuristic and MAC inference), tree decomposition (junction trees and cycle cutset conditioning), and handle Sudoku, map coloring, N-Queens, cryptarithmetic, and timetabling. Adversarial games include Tic-Tac-Toe, Connect Four, Checkers, Othello/Reversi (using Alpha-Beta pruning), and a custom card game (Crazy) with imperfect information, supporting Minimax, Alpha-Beta search, and Monte Carlo Tree Search (MCTS and IS-MCTS).

The framework's core design separates Domains—state representation and problem structure—from Solvers, algorithm implementations. Four base classes (SearchProblem, OptimizationProblem, CSPProblem, GameState) define interfaces; concrete domains inherit from these and implement required methods (e.g., get_actions, heuristic, constraints). This modularity allows any algorithm to run on any compatible domain without modification. A unified interactive CLI launcher (main.py) categorizes 18 demos across Search, Optimization, CSP, and Adversarial Games, with flags to select algorithms (--algo AStar, --algo GeneticAlgorithm, etc.), enable human/AI/human-vs-human modes, and toggle Pygame visualization (--vis flag). All visualizers support dynamic window resizing and standardized HUD controls (SPACE for auto-play, +/– for speed adjustment, LEFT/RIGHT for step control, R for restart).

Benchmarking is automated: python -m benchmarks.run_all_benchmarks --runs 30 executes all algorithm–domain pairs 30 times, saving raw per-run CSV data to results/*.csv. Individual suites (search_benchmark, csp_benchmark, game_benchmark, local_search_benchmark) allow filtering by domain and algorithm. The generate_report command produces high-resolution charts saved to reports/figures/ and markdown summary tables in reports/benchmark_report.md and reports/comparison.md. The project welcomes contributions: new algorithms inherit from SearchAlgorithm and implement search_step(); new CSP heuristics and inference methods are added to csp/heuristics/ or csp/inference/; new game domains inherit from GameState with get_legal_actions(), apply_action(), and is_terminal(); custom visualizers are Pygame classes in visualization/ following standardized HUD patterns. The codebase is licensed under the MIT License and is positioned for educational research, visual learning, and advanced AI algorithm exploration.

Context & Analysis

This release addresses a gap in hands-on AI education: while modern ML frameworks dominate, classical algorithmic AI remains foundational and is often taught through fragmented or closed tutorials. By bundling search, optimization, CSP solving, and game-playing under a single modular architecture, the framework lowers the entry cost for students and researchers who want to experiment without building from scratch.

The design separates domains (problem representations like mazes, TSP, Sudoku, card games) from solvers (algorithms that work across domains), enabling reuse and comparison. The inclusion of interactive visualizers—maze pathfinding, constraint propagation, genetic algorithm populations, game trees—bridges the gap between code and intuition, critical for visual learners. Automated benchmarking pipelines that generate reports and charts further reduce setup friction for comparative evaluation.

The MIT license and explicit invitation for contributions (new algorithms, heuristics, game domains, visualizers) signals intent to grow as a community resource. For educators, researchers prototyping new heuristics, and developers learning AI fundamentals, this consolidates tools that would otherwise require piecing together multiple libraries or writing custom code.

FAQ

What algorithms are included?: The framework supports A*, BFS, DFS, UCS, IDA*, Hill Climbing, Simulated Annealing, Genetic Algorithm, Backtracking with MRV/MAC, Minimax, AlphaBeta, MCTS, and IS-MCTS across search, optimization, constraint satisfaction, and game domains.
How do I run the demos?: Install dependencies (pygame, matplotlib, numpy), then run `python main.py` to launch an interactive CLI menu or `python -m demo.<name> --algo <algorithm> --vis` to run individual demos with visualization.
Can I add my own algorithm or domain?: Yes; the framework is designed for contributions. You can inherit from base classes (SearchAlgorithm, GameState, etc.) in the respective modules, or add new CSP heuristics and custom Pygame visualizers following the standardized patterns.

Hacker NewsRead Original Article