File size: 4,964 Bytes
463f868
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
import os
import sys
import json
import torch
import numpy as np
import random

# Ensure engine_rust is importable
pwd = os.getcwd()
if pwd not in sys.path:
    sys.path.append(pwd)

import engine_rust
from alphazero.vanilla_net import HighFidelityAlphaNet, VanillaTransformerConfig
from alphazero.training.vanilla_action_codec import (
    ACTION_SPACE, 
    policy_id_to_engine_action, 
    build_legal_policy_mask
)

def load_deck_txt(path, db):
    with open(path, "r", encoding="utf-8") as f:
        lines = f.readlines()
    m_list = []
    e_list = []
    for line in lines:
        line = line.strip()
        if not line or line.startswith("#"): continue
        if " x " in line:
            parts = line.split(" x ")
            card_no = parts[0].strip()
            qty = int(parts[1].strip())
            cid = db.id_by_no(card_no)
            if cid is None: continue
            if cid >= 10000: e_list.extend([cid] * qty)
            else: m_list.extend([cid] * qty)
    return {"initial_deck": m_list, "energy": e_list}

def main():
    root = os.getcwd()
    db_path = os.path.join(root, "data", "cards_vanilla.json")
    ckpt_path = os.path.join(root, "checkpoints", "vanilla_overnight", "best.pt")
    deck_path = os.path.join(root, "ai/decks/muse_cup.txt")

    print(f"Loading DB...")
    with open(db_path, "r", encoding="utf-8") as f:
        db_json = f.read()
    rust_db = engine_rust.PyCardDatabase(db_json)
    deck_data = load_deck_txt(deck_path, rust_db)

    print(f"Loading model (preset: tiny)...")
    config = VanillaTransformerConfig.from_preset("tiny")
    model = HighFidelityAlphaNet(config)
    checkpoint = torch.load(ckpt_path, map_location="cpu", weights_only=True)
    model.load_state_dict(checkpoint["model"])
    model.eval()

    num_games = 50
    model_wins = 0
    draws = 0
    total_turns = 0

    print(f"Starting comparison: Model vs Random ({num_games} games)...")

    for i in range(num_games):
        if i % 10 == 0:
            print(f"  Playing game {i}...")
        # Initialize Game
        state = engine_rust.PyGameState(rust_db)
        seed = 42 + i
        state.initialize_game_with_seed(
            deck_data["initial_deck"], deck_data["initial_deck"],
            deck_data["energy"], deck_data["energy"],
            [], [], seed
        )
        state.silent = True
        
        # Determine roles
        # Game 0-24: P0 is Model, P1 is Random
        # Game 25-49: P0 is Random, P1 is Model
        model_player = 0 if i < (num_games // 2) else 1
        
        while not state.is_terminal() and state.turn < 25:
            legal_engine_ids = state.get_legal_action_ids()
            if not legal_engine_ids:
                state.auto_step(rust_db)
                continue
            
            curr_player = state.current_player
            if curr_player == model_player:
                # Model Turn
                obs = state.to_vanilla_tensor()
                obs_t = torch.from_numpy(obs).unsqueeze(0)
                mask = build_legal_policy_mask(state, curr_player, deck_data["initial_deck"], state.phase, legal_engine_ids)
                mask_t = torch.from_numpy(mask).unsqueeze(0)
                
                with torch.no_grad():
                    logits, _ = model(obs_t, mask=mask_t)
                    probs = torch.softmax(logits, dim=1).squeeze(0).numpy()
                
                # Filter strictly by legal engine actions mapping
                legal_probs = []
                legal_actions = []
                for pid in np.where(mask > 0)[0]:
                    eng_id = policy_id_to_engine_action(state, curr_player, pid, state.phase, deck_data["initial_deck"])
                    if eng_id is not None and eng_id in legal_engine_ids:
                        legal_probs.append(probs[pid])
                        legal_actions.append(eng_id)
                
                if not legal_actions:
                    action = random.choice(list(legal_engine_ids))
                else:
                    action = legal_actions[np.argmax(legal_probs)]
            else:
                # Random Turn
                action = random.choice(list(legal_engine_ids))
            
            state.step(int(action))
            state.auto_step(rust_db)
            
        total_turns += state.turn
        winner = state.get_winner()
        if winner == model_player:
            model_wins += 1
        elif winner == -1:
            draws += 1 # Or handle as tie/loss

    print(f"\nResults over {num_games} games:")
    print(f"  Model Wins : {model_wins} ({model_wins/num_games*100:.1f}%)")
    print(f"  Random Wins: {num_games - model_wins - draws}")
    print(f"  Draws      : {draws}")
    print(f"  Avg Turns  : {total_turns/num_games:.1f}")

if __name__ == "__main__":
    main()