from __future__ import annotations

from typing import Any

import torch
import torch.nn as nn


class SIFQLoss(nn.Module):
    """Combine all objective terms with stage-dependent weights."""

    def __init__(
        self,
        matcher_teacher: nn.Module | None,
        sensor_invariance: nn.Module | None,
        degradation_ranking: nn.Module | None,
        orthogonality: nn.Module,
    ):
        super().__init__()
        self.matcher_teacher = matcher_teacher
        self.sensor_invariance = sensor_invariance
        self.degradation_ranking = degradation_ranking
        self.orthogonality = orthogonality

    def forward(
        self,
        outputs: dict[str, torch.Tensor],
        batch: dict[str, Any],
        alpha: float,
        beta: float,
        gamma_stage: float,
    ) -> tuple[torch.Tensor, dict[str, torch.Tensor]]:
        device = outputs["score"].device
        zero = torch.tensor(0.0, device=device)

        l_mat = zero
        if self.matcher_teacher is not None and batch.get("matcher") is not None:
            args = batch["matcher"]
            l_mat = self.matcher_teacher(pred_score=outputs["score"], **args)

        l_sens = zero
        if self.sensor_invariance is not None and batch.get("sensor") is not None:
            args = batch["sensor"]
            l_sens = self.sensor_invariance(**args)

        l_deg = zero
        if self.degradation_ranking is not None and batch.get("degradation") is not None:
            args = batch["degradation"]
            l_deg = self.degradation_ranking(**args)

        l_orth = self.orthogonality(outputs["concepts"])

        total = alpha * l_mat + beta * l_sens + gamma_stage * l_deg + l_orth
        return total, {
            "l_mat": l_mat,
            "l_sens": l_sens,
            "l_deg": l_deg,
            "l_orth": l_orth,
        }