One Piece TCG Deck Optimizer

A PyTorch-based optimization model that generates optimal decklists for the One Piece Trading Card Game using gradient descent and statistical analysis of competitive deck data.

Features

• Differentiable Optimization: Uses Gumbel-Softmax for gradient-based card selection
• Multi-Objective Loss: Balances appearance rates, expected copies, deck size constraints, and diversity
• Data-Driven: Leverages statistical distributions from competitive decklists
• Simulated Annealing: Temperature scheduling for improved convergence

Usage

from Decklist import Decklist
from DeckOptimizer import DeckOptimizer

# Load deck data
decklist = Decklist('bonney_deck_data.csv', winrate=0.55)

# Create optimizer
optimizer = DeckOptimizer(decklist, deck_size=50)

# Generate optimal deck
optimized_deck = optimizer.optimize_deck(
    max_copies=4,
    learning_rate=0.1,
    iterations=1000,
    temperature=2.0,
    anneal_rate=0.995
)

# Display results
optimizer.print_deck(optimized_deck)

How It Works

1. Feature Extraction: Converts card statistics into numerical features
2. Gradient Descent: Optimizes card selection using PyTorch
3. Loss Function:
   • Matches target deck size (50 cards)
   • Maximizes appearance rate of selected cards
   • Aligns with expected copy distributions
   • Penalizes underusing high-appearance cards
   • Encourages deck diversity via entropy
4. Deck Adjustment: Post-processes to ensure exactly 50 cards

Parameters

• max_copies: Maximum copies per card (default: 4)
• learning_rate: Optimization step size (default: 0.1)
• iterations: Training iterations (default: 1000)
• temperature: Initial Gumbel-Softmax temperature (default: 2.0)
• anneal_rate: Temperature decay rate (default: 0.995)