battleship-lgbm

Battleship Agent: LightGBM Model

This repository contains a trained LightGBM model for predicting optimal moves in the game of Battleship. The model uses handcrafted board features to estimate the probability of a hit, enabling intelligent move selection during gameplay.

🔍 Model Overview

• Type: LightGBM binary classifier (LGBMClassifier)
• Task: Predict whether a given cell is a hit (1) or miss (0)
• Training Data: ~27,000 labeled board states with engineered features
• Performance:
  • Accuracy: 71%
  • Hit Recall: 72%
  • Hit F1-score: 0.61

📦 Files

• lgbm_classifier.pkl: Serialized model using joblib
• requirements.txt: Dependencies for loading and using the model
• inference.py (optional): Sample code for loading and using the model

🚀 Usage

from huggingface_hub import hf_hub_download
import joblib

# Load model from Hugging Face
model_path = hf_hub_download(repo_id="your-username/battleship-lgbm-agent", filename="lgbm_classifier.pkl")
model = joblib.load(model_path)

# Predict hit probabilities
y_proba = model.predict_proba(X_board_candidates)[:, 1]
best_idx = y_proba.argmax()
best_move = X_board_candidates.iloc[best_idx]