"""M4 — LAT: Linear Artificial Tomography — SVM decision boundary direction."""

import numpy as np
from sklearn.svm import LinearSVC
from sklearn.preprocessing import StandardScaler

from src.methods.base import SteeringMethod


class LAT(SteeringMethod):
    """LAT — SVM decision boundary as steering direction."""

    @property
    def name(self) -> str:
        return "LAT"

    @property
    def method_id(self) -> str:
        return "M4"

    def extract_vector(
        self,
        h_pos: np.ndarray,
        h_neg: np.ndarray,
        **kwargs,
    ) -> np.ndarray:
        """Compute SVM decision boundary direction.

        Args:
            h_pos: (N_pos, d) positive activations
            h_neg: (N_neg, d) negative activations

        Returns:
            (d,) normal vector to SVM decision boundary
        """
        X = np.concatenate([h_pos, h_neg], axis=0)
        y = np.concatenate([
            np.ones(len(h_pos)),
            np.zeros(len(h_neg)),
        ])

        # Standardise for numeric stability
        scaler = StandardScaler()
        X_scaled = scaler.fit_transform(X)

        C = kwargs.get("C", 1.0)
        max_iter = kwargs.get("max_iter", 5000)

        svm = LinearSVC(C=C, max_iter=max_iter, dual="auto")
        svm.fit(X_scaled, y)

        # Get weight vector in original (unscaled) space
        w = svm.coef_[0] / scaler.scale_
        # Normalise to unit vector
        w = w / (np.linalg.norm(w) + 1e-8)
        return w