nm-ai-astar

Model checkpoints for Challenge 3 (Astar Island) — NM i AI 2026.

Task

Predict probability distributions of terrain types on a 40x40 grid after a 50-year Norse civilization simulation. Scored by entropy-weighted KL divergence.

Models

LightGBM classifiers (one per terrain class, 6 classes total) trained on past round data.

Checkpoint versions

• checkpoints/ — latest production models (v2 retrained)
• checkpoints_v2/ — v2 models with expanded features
• checkpoints_v3/ — v3 models with additional tuning

Terrain classes

Class	Terrain
0	Empty (Ocean/Plains/Empty)
1	Settlement
2	Port
3	Ruin
4	Forest
5	Mountain

Usage

import joblib
import numpy as np

models = {}
for c in range(6):
    models[c] = joblib.load(f"checkpoints/lgbm_class_{c}.joblib")

Team

Agentic Avengers — NM i AI 2026