Spaces:
Running
Running
Upload ai/utils/monitor_self_play.py with huggingface_hub
Browse files- ai/utils/monitor_self_play.py +127 -0
ai/utils/monitor_self_play.py
ADDED
|
@@ -0,0 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json
|
| 2 |
+
import os
|
| 3 |
+
import sys
|
| 4 |
+
|
| 5 |
+
# Add project root to path
|
| 6 |
+
sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
|
| 7 |
+
|
| 8 |
+
import engine_rust
|
| 9 |
+
|
| 10 |
+
from ai.utils.benchmark_decks import parse_deck
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
def monitor_game():
|
| 14 |
+
db_path = "engine/data/cards_compiled.json"
|
| 15 |
+
with open(db_path, "r", encoding="utf-8") as f:
|
| 16 |
+
db_content = f.read()
|
| 17 |
+
db_json = json.loads(db_content)
|
| 18 |
+
db = engine_rust.PyCardDatabase(db_content)
|
| 19 |
+
|
| 20 |
+
# Use simple starter decks
|
| 21 |
+
p_deck = [124, 127] * 20
|
| 22 |
+
p_lives = [1024, 1025, 1027]
|
| 23 |
+
p_energy = [20000] * 10
|
| 24 |
+
|
| 25 |
+
# Setup Agents
|
| 26 |
+
game = engine_rust.PyGameState(db)
|
| 27 |
+
|
| 28 |
+
# Setup Decks (Need correct structure)
|
| 29 |
+
deck0, lives0, energy0 = parse_deck(
|
| 30 |
+
"ai/decks/aqours_cup.txt", db_json["member_db"], db_json["live_db"], db_json.get("energy_db", {})
|
| 31 |
+
)
|
| 32 |
+
deck1, lives1, energy1 = parse_deck(
|
| 33 |
+
"ai/decks/muse_cup.txt", db_json["member_db"], db_json["live_db"], db_json.get("energy_db", {})
|
| 34 |
+
)
|
| 35 |
+
|
| 36 |
+
game.initialize_game(deck0, deck1, energy0, energy1, lives0, lives1)
|
| 37 |
+
|
| 38 |
+
# Setup Agents
|
| 39 |
+
mcts = engine_rust.PyHybridMCTS("ai/models/alphanet.onnx", 0.3)
|
| 40 |
+
|
| 41 |
+
def decode_action(aid, game_p):
|
| 42 |
+
if aid == 0:
|
| 43 |
+
return "Pass/Confirm"
|
| 44 |
+
if 1 <= aid <= 180:
|
| 45 |
+
rel = aid - 1
|
| 46 |
+
h_idx = rel // 3
|
| 47 |
+
s_idx = rel % 3
|
| 48 |
+
if h_idx < len(game_p.hand):
|
| 49 |
+
cid = game_p.hand[h_idx]
|
| 50 |
+
c_name = db_json["member_db"].get(str(cid), {}).get("name", "Unknown")
|
| 51 |
+
return f"Play {c_name} to Slot {s_idx}"
|
| 52 |
+
if 200 <= aid <= 230:
|
| 53 |
+
rel = aid - 200
|
| 54 |
+
s_idx = rel // 10
|
| 55 |
+
return f"Activate Ability on Slot {s_idx}"
|
| 56 |
+
if 300 <= aid <= 360:
|
| 57 |
+
h_idx = aid - 300
|
| 58 |
+
if h_idx < len(game_p.hand):
|
| 59 |
+
cid = game_p.hand[h_idx]
|
| 60 |
+
c_name = db_json["member_db"].get(str(cid), {}).get("name", "Unknown")
|
| 61 |
+
return f"Mulligan Toggle: {c_name}"
|
| 62 |
+
if 400 <= aid <= 460:
|
| 63 |
+
h_idx = aid - 400
|
| 64 |
+
if h_idx < len(game_p.hand):
|
| 65 |
+
cid = game_p.hand[h_idx]
|
| 66 |
+
c_name = db_json["live_db"].get(str(cid), {}).get("name", "Unknown")
|
| 67 |
+
return f"Set Live: {c_name}"
|
| 68 |
+
return f"Unknown({aid})"
|
| 69 |
+
|
| 70 |
+
print("Starting Monitored Game...")
|
| 71 |
+
|
| 72 |
+
step = 0
|
| 73 |
+
while not game.is_terminal() and step < 400: # High limit for trace
|
| 74 |
+
cp_idx = game.current_player
|
| 75 |
+
p0 = game.get_player(0)
|
| 76 |
+
p1 = game.get_player(1)
|
| 77 |
+
cur_p = game.get_player(cp_idx)
|
| 78 |
+
|
| 79 |
+
print(f"\n[Step {step}] Turn {game.turn} | Player {cp_idx} | Phase {game.phase}")
|
| 80 |
+
print(f" P0 Score: {p0.score} | Hand: {len(p0.hand)} | Lives: {len(p0.success_lives)}")
|
| 81 |
+
print(f" P1 Score: {p1.score} | Hand: {len(p1.hand)} | Lives: {len(p1.success_lives)}")
|
| 82 |
+
|
| 83 |
+
is_interactive = game.phase in [-1, 0, 4, 5]
|
| 84 |
+
|
| 85 |
+
if is_interactive:
|
| 86 |
+
# Show Hand Contents
|
| 87 |
+
print(f" Hand Context ({len(cur_p.hand)} cards):")
|
| 88 |
+
for i, cid in enumerate(cur_p.hand):
|
| 89 |
+
m = db_json["member_db"].get(str(cid))
|
| 90 |
+
l = db_json["live_db"].get(str(cid))
|
| 91 |
+
c_name = m["name"] if m else (l["name"] if l else "Unknown")
|
| 92 |
+
c_type = "Member" if m else ("Live" if l else "Energy/Other")
|
| 93 |
+
print(f" [{i}] {c_name} ({c_type}) [ID: {cid}]")
|
| 94 |
+
|
| 95 |
+
legal_ids = game.get_legal_action_ids()
|
| 96 |
+
legal_desc = [decode_action(aid, cur_p) for aid in legal_ids]
|
| 97 |
+
print(f" Legal Actions ({len(legal_ids)}):")
|
| 98 |
+
for aid, desc in zip(legal_ids, legal_desc):
|
| 99 |
+
print(f" - {aid}: {desc}")
|
| 100 |
+
|
| 101 |
+
# Use Hybrid MCTS
|
| 102 |
+
suggestions = mcts.get_suggestions(game, 1000)
|
| 103 |
+
best_action = int(suggestions[0][0])
|
| 104 |
+
|
| 105 |
+
print(f" [Hybrid] Decision: {best_action} ({decode_action(best_action, cur_p)})")
|
| 106 |
+
print(" Probabilities:")
|
| 107 |
+
for a, s, v in suggestions[:5]:
|
| 108 |
+
print(f" - {a} ({decode_action(a, cur_p)}): Score {s:.3f} | Visits {v}")
|
| 109 |
+
|
| 110 |
+
try:
|
| 111 |
+
game.step(best_action)
|
| 112 |
+
except Exception as e:
|
| 113 |
+
print(f"!!! Error stepping {best_action}: {e}")
|
| 114 |
+
break
|
| 115 |
+
else:
|
| 116 |
+
# Auto-step non-interactive phases
|
| 117 |
+
game.step(0)
|
| 118 |
+
|
| 119 |
+
step += 1
|
| 120 |
+
|
| 121 |
+
print("\n--- MONITOR END ---")
|
| 122 |
+
winner = game.get_winner()
|
| 123 |
+
print(f"Final Result: {'Draw' if winner == 2 else f'Player {winner} Wins'} in {step} steps")
|
| 124 |
+
|
| 125 |
+
|
| 126 |
+
if __name__ == "__main__":
|
| 127 |
+
monitor_game()
|