from __future__ import annotations

from dataclasses import dataclass, field
from pathlib import Path


@dataclass
class DataConfig:
    project_root: Path = field(default_factory=lambda: Path(__file__).resolve().parents[1])
    scenario: str = "Non_TMD"
    train_split_name: str = "train"
    val_split_name: str = "val"
    test_split_name: str = "test"
    csv_pattern: str = "*.csv"

    code_column: str = "code"
    time_column: str = "time"
    base_sensor_code: str = "WSMS00012"
    base_axis: str = "value1"
    response_sensor_code: str = "WSMS00007"
    response_axis: str = "value3"

    max_sequence_length: int = 4096
    min_sequence_length: int = 512
    batch_size: int = 8
    num_workers: int = 0
    pin_memory: bool = True
    use_weighted_train_sampler: bool = True
    train_weight_power: float = 1.0
    train_weight_min: float = 0.5
    train_weight_max: float = 8.0
    low_frequency_emphasis_power: float = 1.25
    low_frequency_reference_hz: float = 1.0

    use_steady_state_only: bool = True
    steady_state_start_ratio: float = 0.50
    steady_state_min_samples: int = 256

    interpolation_method: str = "linear"
    normalization_eps: float = 1e-6

    downloads_dir: Path = field(init=False)
    scenario_dir: Path = field(init=False)
    train_dir: Path = field(init=False)
    val_dir: Path = field(init=False)
    test_dir: Path = field(init=False)

    def __post_init__(self) -> None:
        self.project_root = Path(self.project_root).resolve()
        self.downloads_dir = self.project_root / "downloads"
        self.scenario_dir = self.downloads_dir / self.scenario
        self.train_dir = self.scenario_dir / self.train_split_name
        self.val_dir = self.scenario_dir / self.val_split_name
        self.test_dir = self.scenario_dir / self.test_split_name


@dataclass
class ModelConfig:
    input_channels: int = 1
    tcn_channels: tuple[int, ...] = (32, 32, 64, 64)
    kernel_size: int = 7
    dropout: float = 0.15
    dilation_base: int = 2
    pooled_feature_dim: int = 128


@dataclass
class TrainConfig:
    epochs: int = 120
    learning_rate: float = 1e-3
    weight_decay: float = 1e-4
    seed: int = 42
    device: str = "cuda"
    grad_clip_norm: float = 1.0
    use_amp: bool = True
    lr_scheduler_patience: int = 8
    lr_scheduler_factor: float = 0.5
    min_learning_rate: float = 1e-6
    early_stop_patience: int = 15
    checkpoint_dir: str = "checkpoints_rms"
    best_model_name: str = "best_rms_model.pt"
    history_name: str = "training_history.csv"


@dataclass
class LossConfig:
    relative_rms_weight: float = 1.0
    log_rms_weight: float = 0.5
    mae_weight: float = 0.25
    waveform_l1_weight: float = 0.03
    waveform_huber_weight: float = 0.05


@dataclass
class ExperimentConfig:
    data: DataConfig = field(default_factory=DataConfig)
    model: ModelConfig = field(default_factory=ModelConfig)
    train: TrainConfig = field(default_factory=TrainConfig)
    loss: LossConfig = field(default_factory=LossConfig)


def make_rms_forward_config() -> ExperimentConfig:
    return ExperimentConfig()