auto-tetris

An advanced Tetris implementation with AI and human play modes, featuring seamless mode switching and classic gameplay mechanics.

Features

• Dual-mode gameplay: Seamless switching between AI and human control during gameplay
• Classic Tetris mechanics: Line clearing, NES-authentic scoring, progressive difficulty
• Intelligent AI: Multi-factor piece placement evaluation with human-like timing
• Terminal UI: Color-coded pieces, real-time stats, animations, next piece preview

Build and Run

Quick Start

Prerequisites: C compiler, GNU make, ANSI terminal support

$ make        # Build
$ ./tetris    # Run interactive game

Controls:

Key	Action	Mode
Space	Toggle AI/Human mode	Both
P	Pause/unpause game	Both
Q	Quit game	Both
↑	Rotate piece	Human only
←	Move piece left	Human only
→	Move piece right	Human only
↓	Drop piece to bottom	Human only

Game Modes:

• Human: Classic controls with progressive speed increases
• AI: Optimal piece placement with strategic thinking delays

NES Scoring: 1-line: 40×(level+1), 2-line: 100×(level+1), 3-line: 300×(level+1), Tetris: 1200×(level+1)

Benchmark Mode

$ ./tetris -b      # Single game
$ ./tetris -b 10   # Multi-game analysis

Benchmark Metrics:

• Lines Cleared: Total lines cleared before game over
• Score: Final score achieved using standard Tetris scoring
• Pieces Placed: Number of tetromino pieces used
• LCPP: Lines Cleared Per Piece (efficiency metric)
• Game Duration: Time taken to complete the game
• Search Speed: AI decision-making speed (pieces/second)

AI Training

$ make train && ./train    # Basic genetic algorithm training
$ ./train -g 50 -p 12 -e 5 # Custom: 50 generations, 12 population, 5 eval games

Training Options:

Option	Description	Default
-g N	Maximum generations (-1 for infinite)	100
-p N	Population size (2-50)	8
-e N	Evaluation games per individual (1-20)	3
-m RATE	Mutation rate (0.0-1.0)	0.3
-s SEED	Random seed for reproducibility	time-based
-h	Show help and usage	-

Training Process:

• Genetic Evolution: Uses tournament selection, crossover, and mutation
• Fitness Evaluation: Prioritizes lines-cleared-per-piece (LCPP) efficiency
• Progress Tracking: Real-time colored progress bars and fitness statistics
• Weight Export: Automatically saves evolved weights as C header files

TODO

• Add customizable AI difficulty levels
• Create configuration file for key bindings

License

auto-tetris is available under a permissive MIT-style license. Use of this source code is governed by a MIT license that can be found in the LICENSE file.