inference.py

import torch
import numpy as np
import heapq
import os
from config import Config
from model.DNN import DNN
from model.Cube import Cube, TARGET_STATE
from vis import generate_rubiks_html
import time

from solver_utils import *

def generate_html(initial_state, solution_path):
    # 颜色映射 (0=白, 1=红, 2=绿, 3=黄, 4=橙, 5=蓝)
    COLOR_MAP = {
        0: "white",
        1: "red",
        2: "green",
        3: "yellow",
        4: "orange",
        5: "blue"
    }

    # 面顺序和索引范围
    FACE_ORDER = {
        'U': list(range(0, 9)),    # 顶面
        'R': list(range(9, 18)),   # 右面
        'F': list(range(18, 27)),  # 前面
        'D': list(range(27, 36)),  # 底面
        'L': list(range(36, 45)),  # 左面
        'B': list(range(45, 54))   # 后面
    }

    # 将初始状态转换为generate_rubiks_html需要的格式
    initial_state_dict = {}
    for face, indices in FACE_ORDER.items():
        initial_state_dict[face] = [COLOR_MAP[initial_state[i]] for i in indices]

    # 生成每一步的状态
    moves = []
    for state in solution_path:
        move_state = {}
        for face, indices in FACE_ORDER.items():
            move_state[face] = [COLOR_MAP[state[i]] for i in indices]
        moves.append(move_state)
        
    # 调用generate_rubiks_html生成网页
    output_file = "rubiks_solution.html"
    generate_rubiks_html(initial_state_dict, FACE_ORDER, moves, output_file)
    print(f"已生成解决方案网页: {output_file}")


def main():
    # 加载配置
    config = Config()
    args = config.parse_args()

    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

    # 加载模型
    model = load_model(args.model_path, device=device)

    # 创建Cube对象
    cube = Cube()

    # 初始状态为目标状态
    initial_state = TARGET_STATE.copy()
    state_path = []

    if args.actions:
        # 使用用户指定的动作序列
        print(f"使用指定的动作序列: {args.actions}")
        shuffle_actions = args.actions.split()
        # 验证动作是否有效
        valid_actions = set(cube.moves.keys())
        invalid_actions = [action for action in shuffle_actions if action not in valid_actions]
        if invalid_actions:
            raise ValueError(f"无效的动作: {invalid_actions}，有效的动作是: {valid_actions}")
    else:
        # 随机生成动作序列
        print("随机打乱魔方100次...")
        all_actions = list(cube.moves.keys())
        shuffle_actions = np.random.choice(all_actions, size=100, replace=True)

    # 应用动作序列
    for action in shuffle_actions:
        initial_state = cube.apply_action(initial_state, action)
        state_path.append(initial_state.copy())

    print("开始A*搜索...")

    # 执行A*搜索
    start_time = time.time()
    action_path, solution_state_path = a_star_search(initial_state, model, cube)
    end_time = time.time()
    solving_time = end_time - start_time

    # 保存为可视化结果
    if solution_state_path:
        # 合并打乱路径和解决方案路径以展示完整过程
        # full_state_path = state_path # + solution_state_path[1:]
        generate_html(initial_state.copy(), solution_state_path)
        # generate_html(initial_state, solution_state_path[1:])
        print(f"找到解决方案，路径长度: {len(solution_state_path)}")
        print(f"求解时间: {solving_time:.4f} 秒")  # 打印求解时间
        print("打乱路径:", shuffle_actions)
        if action_path:
            print("解决方案路径:", action_path)
    else:
        print("未找到解决方案")

if __name__ == '__main__':
    main()