class IntegratedGradientsExplainer(FeatureAttributionExplainer):
    """Integrated Gradients explainer for computing path-integrated attributions.

    This explainer computes attributions using Integrated Gradients, which integrates
    gradients along a straight path from a baseline input to the actual input.
    This method satisfies important axioms including sensitivity and implementation
    invariance, making it a robust attribution method. Supports both single-target
    and multi-target modes for both single-target and multi-target scenarios.

    The Integrated Gradients method provides theoretically grounded attributions
    by computing the integral of gradients along the path from baseline to input.

    Args:
        model: The PyTorch model whose output is to be explained.
        multi_target: Whether to use multi-target mode. When True, can compute
            attributions for multiple targets simultaneously. Defaults to False.
        internal_batch_size: Batch size for internal computations. Defaults to 64.
        grad_batch_size: Batch size for gradient computations. Defaults to 64.
        n_steps: Number of steps for the integral approximation. Defaults to 50.

    Examples:
        Single-target usage:
        >>> import torch
        >>> from torchxai.data_types import SingleTargetAcrossBatch
        >>>
        >>> model = torch.nn.Linear(10, 2)
        >>> explainer = IntegratedGradientsExplainer(model)
        >>> inputs   = torch.randn(1, 10)
        >>> baseline = torch.zeros(1, 10)
        >>> attributions = explainer.explain(
        ...     inputs=inputs,
        ...     baselines=baseline,
        ...     target=SingleTargetAcrossBatch(index=0),
        ... )
        >>> attributions.shape   # (1, 10)

        Multi-target usage:
        >>> explainer_mt = IntegratedGradientsExplainer(model, multi_target=True)
        >>> mt_attributions = explainer_mt.explain(
        ...     inputs=inputs,
        ...     baselines=baseline,
        ...     target=[SingleTargetAcrossBatch(index=0), SingleTargetAcrossBatch(index=1)],
        ... )
        >>> len(mt_attributions), mt_attributions[0].shape   # 2, (1, 10)
    """

    __repr_attrs__ = [
        "_multi_target",
        "_internal_batch_size",
        "_grad_batch_size",
        "n_steps",
        "return_convergence_delta",
    ]

    def __init__(
        self,
        model: Module,
        multi_target: bool = False,
        internal_batch_size: int = 50,
        grad_batch_size: int = 64,
        n_steps: int = 50,
        return_convergence_delta: bool = False,
    ) -> None:
        """Initialize the IntegratedGradientsExplainer.

        Args:
            model: The model whose output is to be explained.
            multi_target: Whether to use multi-target mode. Defaults to False.
            internal_batch_size: Batch size for internal computations. Defaults to 50.
            grad_batch_size: Batch size for gradient computations. Defaults to 64.
            n_steps: Number of steps for the integral approximation. Defaults to 50.
            return_convergence_delta: Whether to return convergence delta for
                completeness check. Defaults to False.
        """
        self.n_steps = n_steps
        self.return_convergence_delta = return_convergence_delta

        super().__init__(
            model, multi_target, internal_batch_size, grad_batch_size=grad_batch_size
        )

    def _init_single_target_explanation_fn(self) -> Callable:
        """Initialize single-target Integrated Gradients attribution function.

        Returns:
            Captum IntegratedGradients attribution function for single targets.
        """
        return partial(
            IntegratedGradients(self._model).attribute,
            n_steps=self.n_steps,
            return_convergence_delta=self.return_convergence_delta,
            internal_batch_size=self._internal_batch_size,
        )  # type: ignore

    def _init_multi_target_explanation_fn(self) -> Callable:
        """Initialize multi-target Integrated Gradients attribution function.

        Returns:
            MultiTargetIntegratedGradients attribution function for multiple targets.
        """
        return partial(
            MultiTargetIntegratedGradients(
                self._model, grad_batch_size=self._grad_batch_size
            ).attribute,
            n_steps=self.n_steps,
            return_convergence_delta=self.return_convergence_delta,
            internal_batch_size=self._internal_batch_size,
        )

    def explain(
        self,
        inputs: TensorOrTupleOfTensorsGeneric,
        target: ExplanationTargetType | list[ExplanationTargetType],
        baselines: TensorOrTupleOfTensorsGeneric | None = None,
        additional_forward_args: tuple[Any, ...] | None = None,
    ) -> TensorOrTupleOfTensorsGeneric | list[TensorOrTupleOfTensorsGeneric]:
        """Compute Integrated Gradients attributions for the given inputs.

        Args:
            inputs: Input tensor(s) for attribution computation.
            target: An `ExplanationTargetType` (e.g. `SingleTargetAcrossBatch`) for single-target
                mode, or a list of them for multi-target mode.
            baselines: Baseline tensors representing reference values. If None,
                uses zero baselines. Should match the structure of inputs.
            additional_forward_args: Additional arguments for model forward pass.

        Returns:
            Tensor in single-target mode. List of Tensors, one per target, in multi-target mode.

            If return_convergence_delta was set to True during initialization, the return
            format may include convergence delta information depending on the underlying
            implementation.

        Note:
            The number of integration steps and convergence delta behavior are controlled
            by parameters set during initialization (n_steps and return_convergence_delta).
            More steps generally provide more accurate approximations but increase computation time.

        Examples:
            >>> # With convergence delta enabled at initialization
            >>> explainer = IntegratedGradientsExplainer(
            ...     model, n_steps=100, return_convergence_delta=True
            ... )
            >>> attributions = explainer.explain(
            ...     inputs=OrderedDict({"input": torch.randn(2, 10)}),
            ...     target=torch.tensor([0, 1]),
            ...     baselines=OrderedDict({"input": torch.zeros(2, 10)}),
            ... )
            >>>
            >>> # With automatic zero baselines
            >>> attributions = explainer.explain(
            ...     inputs=torch.randn(2, 10), target=torch.tensor([0, 1])
            ... )
        """
        return self._default_explain(
            inputs=inputs,
            target=target,
            baselines=baselines,
            additional_forward_args=additional_forward_args,
        )