"""
全局配置 - 验证码识别多模型系统 (CaptchaBreaker)

定义字符集、图片尺寸、路径、训练超参等所有全局常量。
"""

import os
from pathlib import Path

# ============================================================
# 项目根目录
# ============================================================
PROJECT_ROOT = Path(__file__).resolve().parent

# ============================================================
# 数据目录
# ============================================================
DATA_DIR = PROJECT_ROOT / "data"
SYNTHETIC_DIR = DATA_DIR / "synthetic"
REAL_DIR = DATA_DIR / "real"
CLASSIFIER_DIR = DATA_DIR / "classifier"

# 合成数据子目录
SYNTHETIC_NORMAL_DIR = SYNTHETIC_DIR / "normal"
SYNTHETIC_MATH_DIR = SYNTHETIC_DIR / "math"
SYNTHETIC_3D_DIR = SYNTHETIC_DIR / "3d"

# 真实数据子目录
REAL_NORMAL_DIR = REAL_DIR / "normal"
REAL_MATH_DIR = REAL_DIR / "math"
REAL_3D_DIR = REAL_DIR / "3d"

# ============================================================
# 模型输出目录
# ============================================================
CHECKPOINTS_DIR = PROJECT_ROOT / "checkpoints"
ONNX_DIR = PROJECT_ROOT / "onnx_models"

# 确保关键目录存在
for _dir in [
    SYNTHETIC_NORMAL_DIR, SYNTHETIC_MATH_DIR, SYNTHETIC_3D_DIR,
    REAL_NORMAL_DIR, REAL_MATH_DIR, REAL_3D_DIR,
    CLASSIFIER_DIR, CHECKPOINTS_DIR, ONNX_DIR,
]:
    _dir.mkdir(parents=True, exist_ok=True)

# ============================================================
# 字符集定义
# ============================================================
# 普通字符验证码: 按当前本地配置保留易混淆字符，覆盖完整数字 + 大写字母
NORMAL_CHARS = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"

# 算式验证码: 数字 + 运算符
MATH_CHARS = "0123456789+-×÷=?"

# 3D 验证码: 继续使用去掉易混淆字符的精简字符集
THREED_CHARS = "23456789ABCDEFGHJKMNPQRSTUVWXYZ"

# 验证码类型列表 (调度分类器输出)
CAPTCHA_TYPES = ["normal", "math", "3d"]
NUM_CAPTCHA_TYPES = len(CAPTCHA_TYPES)

# ============================================================
# 图片尺寸配置 (H, W)
# ============================================================
IMAGE_SIZE = {
    "classifier": (64, 128),     # 调度分类器输入
    "normal":     (40, 120),     # 普通字符识别
    "math":       (40, 160),     # 算式识别 (更宽)
    "3d":         (60, 160),     # 3D 立体识别
}

# ============================================================
# 验证码生成参数
# ============================================================
GENERATE_CONFIG = {
    "normal": {
        "char_count_range": (4, 5),           # 字符数量: 4-5 个
        "bg_color_range": (230, 255),         # 浅色背景 RGB 各通道
        "rotation_range": (-15, 15),          # 字符旋转角度
        "noise_line_range": (2, 5),           # 干扰线数量
        "noise_point_num": 100,               # 噪点数量
        "blur_radius": 0.8,                   # 高斯模糊半径
        "image_size": (120, 40),              # 生成图片尺寸 (W, H)
    },
    "math": {
        "operand_range": (1, 30),             # 操作数范围
        "operators": ["+", "-", "×"],         # 支持的运算符 (除法只生成能整除的)
        "image_size": (160, 40),              # 生成图片尺寸 (W, H)
        "bg_color_range": (230, 255),
        "rotation_range": (-10, 10),
        "noise_line_range": (2, 4),
    },
    "3d": {
        "char_count_range": (4, 5),
        "image_size": (160, 60),              # 生成图片尺寸 (W, H)
        "shadow_offset": (3, 3),              # 阴影偏移
        "perspective_intensity": 0.3,         # 透视变换强度
    },
}

# ============================================================
# 训练配置
# ============================================================
TRAIN_CONFIG = {
    "classifier": {
        "epochs": 30,
        "batch_size": 128,
        "lr": 1e-3,
        "scheduler": "cosine",
        "synthetic_samples": 30000,           # 每类 10000
        "val_split": 0.1,                     # 验证集比例
    },
    "normal": {
        "epochs": 50,
        "batch_size": 128,
        "lr": 1e-3,
        "scheduler": "cosine",
        "synthetic_samples": 60000,
        "loss": "CTCLoss",
        "val_split": 0.1,
    },
    "math": {
        "epochs": 50,
        "batch_size": 128,
        "lr": 1e-3,
        "scheduler": "cosine",
        "synthetic_samples": 60000,
        "loss": "CTCLoss",
        "val_split": 0.1,
    },
    "threed": {
        "epochs": 80,
        "batch_size": 64,
        "lr": 5e-4,
        "scheduler": "cosine",
        "synthetic_samples": 80000,
        "loss": "CTCLoss",
        "val_split": 0.1,
    },
}

# ============================================================
# 数据增强参数 (训练时使用)
# ============================================================
AUGMENT_CONFIG = {
    "degrees": 8,                             # RandomAffine 旋转范围
    "translate": (0.05, 0.05),                # 平移范围
    "scale": (0.95, 1.05),                    # 缩放范围
    "brightness": 0.3,                        # ColorJitter 亮度
    "contrast": 0.3,                          # ColorJitter 对比度
    "blur_kernel": 3,                         # GaussianBlur 核大小
    "blur_sigma": (0.1, 0.5),                 # GaussianBlur sigma
    "erasing_prob": 0.15,                     # RandomErasing 概率
    "erasing_scale": (0.01, 0.05),            # RandomErasing 面积比
}

# ============================================================
# ONNX 导出配置
# ============================================================
ONNX_CONFIG = {
    "opset_version": 18,
    "dynamic_batch": True,                    # 支持动态 batch size
}

# ============================================================
# 推理配置
# ============================================================
INFERENCE_CONFIG = {
    "default_models_dir": str(ONNX_DIR),
    "normalize_mean": 0.5,
    "normalize_std": 0.5,
}

# ============================================================
# 随机种子 (保证数据生成可复现)
# ============================================================
RANDOM_SEED = 42

# ============================================================
# 设备配置 (优先 GPU，回退 CPU)
# 延迟导入 torch，避免仅使用生成器时必须安装 torch
# ============================================================
def get_device():
    """返回可用的 torch 设备，优先 GPU。"""
    import torch
    return torch.device("cuda" if torch.cuda.is_available() else "cpu")

# ============================================================
# 服务配置 (可选 HTTP 服务)
# ============================================================
SERVER_CONFIG = {
    "host": "0.0.0.0",
    "port": 8080,
}