"""Cluster bootstrap for accuracies and paired differences.

Items nested in generating templates (clusters) are not independent; the bootstrap resamples
clusters with replacement. Paired differences use identical items for both conditions.
"""
from __future__ import annotations

from dataclasses import dataclass

import numpy as np


@dataclass
class CI:
    estimate: float
    low: float
    high: float
    level: float
    n_items: int
    n_clusters: int
    n_boot: int

    def to_dict(self) -> dict:
        return self.__dict__.copy()


def _cluster_index(clusters: list[str]) -> tuple[np.ndarray, int]:
    uniq = {c: i for i, c in enumerate(dict.fromkeys(clusters))}
    return np.array([uniq[c] for c in clusters]), len(uniq)


def cluster_bootstrap_mean(values: np.ndarray, clusters: list[str], n_boot: int = 2000, level: float = 0.95, seed: int = 0) -> CI:
    values = np.asarray(values, dtype=float)
    cid, k = _cluster_index(clusters)
    rng = np.random.default_rng(seed)
    sums = np.bincount(cid, weights=values, minlength=k)
    counts = np.bincount(cid, minlength=k).astype(float)
    stats = np.empty(n_boot)
    for b in range(n_boot):
        draw = rng.integers(0, k, k)
        stats[b] = sums[draw].sum() / counts[draw].sum()
    a = (1 - level) / 2
    return CI(float(values.mean()), float(np.quantile(stats, a)), float(np.quantile(stats, 1 - a)), level, len(values), k, n_boot)


def cluster_bootstrap_paired_diff(a: np.ndarray, b: np.ndarray, clusters: list[str], n_boot: int = 2000, level: float = 0.95, seed: int = 0) -> CI:
    """CI for mean(a) - mean(b) on identical items (paired), resampling clusters."""
    return cluster_bootstrap_mean(np.asarray(a, float) - np.asarray(b, float), clusters, n_boot, level, seed)


def seeds_summary(values: list[float]) -> dict:
    v = np.asarray(values, float)
    return {"n_seeds": int(len(v)), "mean": float(v.mean()), "sd": float(v.std(ddof=1)) if len(v) > 1 else None,
            "min": float(v.min()), "max": float(v.max())}