🧠 Hex Conquest — ML Training Dashboard

Loading...

📈 Episode Reward

📉 Training Loss

💰 Max Gold Production/tick

đŸ—ș Territory & Buildings

🎯 Epsilon (Exploration)

🚀 Missile Silo Rate (rolling 100)

🎬 Action Distribution (last episode)

📖 Reference — Actions & Metrics

🎼 Actions (14 active)

NOTHINGDo nothing this tick
CLAIMClaim an adjacent unclaimed tile
DIG_OWNDig an owned mountain tile deeper
DIG_ADJDig an adjacent mountain tile
SOLARBuild Solar Panel (☀ energy producer)
WINDBuild Wind Turbine (💹 constant energy)
BATTERYBuild Battery (🔋 energy storage)
FACTORYBuild Factory (🏭 converts energy → gold)
DRILLBuild Drill Silo (🔹 digs remote tiles)
MISSILEBuild Missile Silo (🚀 damages enemy buildings)
SHIELDBuild Shield (đŸ›Ąïž protects tile from missiles)
UPGRADEUpgrade best building (higher level = more output)
FIRE_DRILLFire a drill at a remote tile
FIRE_MISSFire a missile at enemy tile
14-17ALLOC actions DISABLED — fixed [1,1,1] balanced

📊 Metrics

Episode RewardTotal reward accumulated during episode
Training LossTraining loss (lower = more stable learning)
GoldGold at end of episode (đŸȘ™)
TerritoryTiles owned at end of episode
BuildingsTotal buildings placed
EpsilonExploration rate (1.0=random → 0.05=greedy)
Missile Rate% of recent episodes with a missile silo built

🧠 Reward V7 (Simplified)

-0.3DO_NOTHING (flat penalty)
+1.0CLAIM tile initiated
+0.5DIG (own or adjacent)
+5.0BUILD any building
+5.0UPGRADE building
+2.0FIRE_DRILL
+10Missile hits enemy building
+20Missile destroys enemy building
+2.0/tileTerritory gained (continuous)
+0.05/goldGold production (continuous)
+500Victory / -500 Defeat
⚙Max 2 concurrent constructions
⚙Episodes: 1500 ticks