---
library_name: stable-baselines3
tags:
  - reinforcement-learning
  - battleship
  - ppo
  - maskable-ppo
  - sb3-contrib
---

# Battleship PPO Agent — 106.9 mean shots

Best checkpoint from Mar 21 autoresearch session (50 experiments).

## Key Result
| Metric | Value |
|--------|-------|
| eval/mean_shots | **106.9** |
| previous_best | 109.6 |
| improvement | 2.7 shots |

## Training Config
| Parameter | Value |
|-----------|-------|
| lr_schedule | linear |
| n_epochs | 2 |
| lr | 1e-5 |
| clip_range | 0.05 |
| ent_coef | 0.002 |
| ent_coef_final | 0.0002 |
| n_envs | 4 |
| n_steps | 128 |
| batch_size | 128 |
| gamma | 0.995 |
| reward_preset | efficiency_v2_infogain |
| seed | 51 |
| device | cuda (RTX 5070 Ti) |
| time_budget | 30min |