>_ Train RL Hunter
DQN · runs in browser · saves to game automatically
← back to escape grid
// status
state
idle
episode
0 / 8000
win rate (last 200)
—
avg reward
—
epsilon
1.000
replay buffer
0 / 60000
// controls
episodes
8000
speed (ep/tick)
3
▶ Start Training
✓ weights saved — H: RL ready in game
⬇ Save Weights to Game
⬇ Download hunter_rl.json
↺ Reset
// hyperparams
network
62→128→128→14
optimizer
Adam lr=1e-3
gamma
0.95
batch
64
ε decay
×0.9995/ep
target sync
every 300 steps
participant AI
70% greedy
// win rate over time
// training log