from __future__ import annotations

import html
import math
from dataclasses import dataclass
from pathlib import Path
from typing import Iterable

import pandas as pd


DATA_PATH = Path(__file__).parent / "data" / "leaderboard.csv"

METRIC_COLUMNS = [
    "Vis",
    "Aud (PQ)",
    "AV",
    "Lip",
    "Text",
    "Face",
    "Music",
    "Speech",
    "Lo-Phy",
    "Hi-Phy",
    "Holistic",
    "Total",
]

SORT_COLUMNS = ["Rank", "Model", "Components", "Component Type", *METRIC_COLUMNS]
SORT_CHOICES = [
    ("Rank", "Rank"),
    ("Model", "Model"),
    ("Components", "Components"),
    ("Type", "Component Type"),
    *[(metric, metric) for metric in METRIC_COLUMNS],
]

LOWER_IS_BETTER = {"AV", "Lip"}

NUMERIC_COLUMNS = METRIC_COLUMNS

FORMATTERS = {
    "Vis": "{:.3f}",
    "Aud (PQ)": "{:.2f}",
    "AV": "{:.2f}",
    "Lip": "{:.2f}",
    "Text": "{:.2f}",
    "Face": "{:.2f}",
    "Music": "{:.2f}",
    "Speech": "{:.2f}",
    "Lo-Phy": "{:.2f}",
    "Hi-Phy": "{:.2f}",
    "Holistic": "{:.2f}",
    "Total": "{:.2f}",
}

GROUP_WEIGHTS = {
    "Basic Uni-modal": 0.2,
    "Basic Cross-modal": 0.2,
    "Fine-grained": 0.6,
}

GROUP_DIMENSIONS = {
    "Basic Uni-modal": ["Vis", "Aud (PQ)"],
    "Basic Cross-modal": ["AV", "Lip"],
    "Fine-grained": ["Text", "Face", "Music", "Speech", "Lo-Phy", "Hi-Phy", "Holistic"],
}


@dataclass(frozen=True)
class Standing:
    best: float | None
    second: float | None


def load_leaderboard(path: Path = DATA_PATH) -> pd.DataFrame:
    df = pd.read_csv(path)
    for column in NUMERIC_COLUMNS:
        df[column] = pd.to_numeric(df[column], errors="coerce")
    df = df.sort_values("Total", ascending=False, na_position="last").reset_index(drop=True)
    df.insert(0, "Rank", range(1, len(df) + 1))
    return df


def filter_leaderboard(
    df: pd.DataFrame,
    component_type: str = "All",
    query: str = "",
    sort_by: str = "Total",
    sort_order: str = "Descending",
) -> pd.DataFrame:
    view = df.copy()

    if component_type != "All":
        view = view[view["Component Type"] == component_type]

    query = query.strip().lower()
    if query:
        mask = (
            view["Model"].str.lower().str.contains(query, regex=False)
            | view["Components"].str.lower().str.contains(query, regex=False)
        )
        view = view[mask]

    sort_by = _normalize_sort_column(sort_by)
    ascending = _is_ascending_sort(sort_by, sort_order)

    if sort_by in view.columns:
        sort_kwargs = {
            "ascending": ascending,
            "na_position": "last",
            "kind": "mergesort",
        }
        if sort_by in {"Model", "Components", "Component Type"}:
            sort_kwargs["key"] = lambda column: column.astype(str).str.casefold()
        view = view.sort_values(sort_by, **sort_kwargs)

    return view.reset_index(drop=True)


def metric_standings(df: pd.DataFrame) -> dict[str, Standing]:
    standings: dict[str, Standing] = {}
    for metric in METRIC_COLUMNS:
        values = sorted(
            {float(v) for v in df[metric].dropna()},
            reverse=metric not in LOWER_IS_BETTER,
        )
        standings[metric] = Standing(
            best=values[0] if values else None,
            second=values[1] if len(values) > 1 else None,
        )
    return standings


def normalized_score(metric: str, value: float) -> float:
    if metric == "Vis":
        return _clamp(value * 100.0, 0.0, 100.0)
    if metric == "Aud (PQ)":
        return _clamp(value * 10.0, 0.0, 100.0)
    if metric == "AV":
        return _clamp(100.0 * (1.0 - value / 0.5), 0.0, 100.0)
    if metric == "Lip":
        return _clamp(100.0 * (1.0 - value / 8.0), 0.0, 100.0)
    if metric == "Lo-Phy":
        return _clamp(value * 20.0, 0.0, 100.0)
    return _clamp(value, 0.0, 100.0)


def compute_total_from_metrics(row: pd.Series) -> float:
    group_scores: list[float] = []
    group_weights: list[float] = []

    for group_name, metrics in GROUP_DIMENSIONS.items():
        values = []
        for metric in metrics:
            value = row.get(metric)
            if pd.isna(value):
                continue
            values.append(normalized_score(metric, float(value)))
        if values:
            group_scores.append(sum(values) / len(values))
            group_weights.append(GROUP_WEIGHTS[group_name])

    if not group_scores:
        return float("nan")

    weighted = sum(score * weight for score, weight in zip(group_scores, group_weights))
    return weighted / sum(group_weights)


def render_summary(df: pd.DataFrame, view: pd.DataFrame) -> str:
    top = df.sort_values("Total", ascending=False).iloc[0]
    open_source = df[df["Component Type"] == "Open-source"].sort_values("Total", ascending=False)
    best_open = open_source.iloc[0] if len(open_source) else None
    best_av = df.sort_values("AV", ascending=True).iloc[0]
    best_speech = df.sort_values("Speech", ascending=False).iloc[0]

    cards = [
        _summary_card("Models", f"{len(view)} / {len(df)}", "shown in current view"),
        _summary_card("Top Total", _score(top["Total"]), str(top["Model"])),
        _summary_card(
            "Best Open-source",
            _score(best_open["Total"]) if best_open is not None else "NA",
            str(best_open["Model"]) if best_open is not None else "No entry",
        ),
        _summary_card("Lowest AV Offset", _score(best_av["AV"]), str(best_av["Model"])),
        _summary_card("Highest Speech", _score(best_speech["Speech"]), str(best_speech["Model"])),
    ]
    return '<div class="summary-grid">' + "".join(cards) + "</div>"


def render_table(
    df: pd.DataFrame,
    standings: dict[str, Standing],
    sort_by: str = "Total",
    sort_order: str = "Descending",
) -> str:
    if df.empty:
        return '<div class="empty-state">No matching models.</div>'

    sort_by = _normalize_sort_column(sort_by)
    headers = [
        ("Rank", "Rank"),
        ("Model", "Model"),
        ("Components", "Components"),
        ("Type", "Component Type"),
        *[(metric, metric) for metric in METRIC_COLUMNS],
    ]
    header_html = "".join(_header_cell(label, column, sort_by, sort_order) for label, column in headers)

    rows = []
    for _, row in df.iterrows():
        cells = [
            f'<td class="rank-cell">#{int(row["Rank"])}</td>',
            f'<td class="model-cell">{html.escape(str(row["Model"]))}</td>',
            f'<td class="components-cell">{render_component_badges(str(row["Components"]))}</td>',
            f'<td>{_type_badge(str(row["Component Type"]))}</td>',
        ]
        for metric in METRIC_COLUMNS:
            cells.append(_metric_cell(metric, row[metric], standings[metric]))
        rows.append("<tr>" + "".join(cells) + "</tr>")

    return (
        '<div class="table-shell"><table class="leaderboard-table">'
        f"<thead><tr>{header_html}</tr></thead><tbody>{''.join(rows)}</tbody>"
        "</table></div>"
    )


def render_component_badges(components: str) -> str:
    badges = []
    for component in _split_components(components):
        lowered = component.lower()
        kind = "proprietary" if "proprietary" in lowered else "open" if "open-source" in lowered else "neutral"
        label = component.replace(" (Proprietary)", "").replace(" (Open-source)", "")
        badges.append(f'<span class="component-badge {kind}">{html.escape(label)}</span>')
    return "".join(badges)


def render_profile(df: pd.DataFrame, model: str) -> str:
    if df.empty:
        return ""
    if not model or model not in set(df["Model"]):
        model = str(df.sort_values("Total", ascending=False).iloc[0]["Model"])

    row = df[df["Model"] == model].iloc[0]
    metric_blocks = []
    for metric in METRIC_COLUMNS[:-1]:
        value = float(row[metric])
        normalized = normalized_score(metric, value)
        direction = "lower is better" if metric in LOWER_IS_BETTER else "higher is better"
        metric_blocks.append(
            f"""
            <div class="profile-metric">
              <div class="profile-metric-head">
                <span>{html.escape(metric)}</span>
                <strong>{_format_metric(metric, value)}</strong>
              </div>
              <div class="bar-track"><div class="bar-fill" style="width: {normalized:.1f}%"></div></div>
              <small>{normalized:.1f} normalized, {direction}</small>
            </div>
            """
        )

    return f"""
    <div class="profile-panel">
      <div>
        <p class="eyebrow">Model profile</p>
        <h2>{html.escape(str(row["Model"]))}</h2>
        <div class="profile-components">{render_component_badges(str(row["Components"]))}</div>
      </div>
      <div class="profile-total">
        <span>Total</span>
        <strong>{_score(row["Total"])}</strong>
      </div>
      <div class="profile-grid">{''.join(metric_blocks)}</div>
    </div>
    """


def render_methodology() -> str:
    groups = "".join(
        f"""
        <div class="method-card">
          <h3>{html.escape(name)}</h3>
          <strong>{weight:.1f}</strong>
          <p>{html.escape(', '.join(metrics))}</p>
        </div>
        """
        for name, weight in GROUP_WEIGHTS.items()
        for metrics in [GROUP_DIMENSIONS[name]]
    )
    return f"""
    <div class="methodology">
      <div class="method-grid">{groups}</div>
      <p>
        Total uses AVGen-Bench Scheme 2: group-weighted normalized metrics with
        Vis x 100, Aud(PQ) x 10, Lo-Phy x 20, AV = 100 * max(0, 1 - AV / 0.5),
        Lip = 100 * max(0, 1 - Lip / 8), and the remaining metrics already on
        a 0-100 scale.
      </p>
    </div>
    """


def model_choices(df: pd.DataFrame) -> list[str]:
    return list(df.sort_values("Total", ascending=False)["Model"])


def _metric_cell(metric: str, value: float, standing: Standing) -> str:
    if pd.isna(value):
        return '<td class="metric-cell muted">NA</td>'
    numeric = float(value)
    classes = ["metric-cell"]
    if _close(numeric, standing.best):
        classes.append("best")
    elif _close(numeric, standing.second):
        classes.append("second")
    return f'<td class="{" ".join(classes)}">{_format_metric(metric, numeric)}</td>'


def _type_badge(component_type: str) -> str:
    kind = component_type.lower().replace("-", "").replace(" ", "")
    return f'<span class="type-badge {html.escape(kind)}">{html.escape(component_type)}</span>'


def _summary_card(label: str, value: str, detail: str) -> str:
    return f"""
    <div class="summary-card">
      <span>{html.escape(label)}</span>
      <strong>{html.escape(value)}</strong>
      <small>{html.escape(detail)}</small>
    </div>
    """


def _header_cell(label: str, column: str, sort_by: str, sort_order: str) -> str:
    if column != sort_by:
        return f"<th>{html.escape(label)}</th>"

    direction = "ascending" if _is_ascending_sort(sort_by, sort_order) else "descending"
    indicator = "&uarr;" if direction == "ascending" else "&darr;"
    return (
        f'<th class="sorted" aria-sort="{direction}">'
        f"{html.escape(label)}"
        f'<span class="sort-indicator" aria-hidden="true">{indicator}</span>'
        "</th>"
    )


def _normalize_sort_column(sort_by: str) -> str:
    if sort_by == "Type":
        return "Component Type"
    return sort_by if sort_by in SORT_COLUMNS else "Total"


def _is_ascending_sort(sort_by: str, sort_order: str) -> bool:
    if sort_order == "Best first":
        return sort_by in LOWER_IS_BETTER
    return sort_order == "Ascending"


def _split_components(value: str) -> Iterable[str]:
    return [part.strip() for part in value.split("|") if part.strip()]


def _format_metric(metric: str, value: float) -> str:
    return FORMATTERS[metric].format(float(value))


def _score(value: float) -> str:
    if pd.isna(value):
        return "NA"
    return f"{float(value):.2f}"


def _close(left: float, right: float | None) -> bool:
    if right is None:
        return False
    return math.isclose(float(left), float(right), rel_tol=0.0, abs_tol=1e-9)


def _clamp(value: float, low: float, high: float) -> float:
    return max(low, min(high, value))