"""Feature engineering for prediction market analysis."""

from __future__ import annotations

import numpy as np
import pandas as pd


def compute_log_returns(prices: pd.Series) -> pd.Series:
    return np.log(prices / prices.shift(1))


def compute_realized_volatility(returns: pd.Series, window: int = 5) -> pd.Series:
    return (returns ** 2).rolling(window).sum().apply(np.sqrt)


def compute_probability_changes(probs: pd.DataFrame) -> pd.DataFrame:
    return probs.diff()


def compute_roll_spread(prices: pd.Series, window: int = 20) -> pd.Series:
    """Roll (1984) bid-ask spread estimator."""
    dp = prices.diff()

    def _roll_cov(x: np.ndarray) -> float:
        if len(x) < 3:
            return np.nan
        return np.cov(x[:-1], x[1:])[0, 1]

    cov = dp.rolling(window).apply(_roll_cov, raw=True)
    return 2 * np.sqrt(np.maximum(-cov, 0))


def compute_amihud_illiquidity(returns: pd.Series, volume: pd.Series) -> pd.Series:
    return (returns.abs() / volume).replace([np.inf, -np.inf], np.nan)


def compute_abnormal_volume(volume: pd.Series, window: int = 20) -> pd.Series:
    return volume / volume.rolling(window).mean()


def engineer_features(prices_df: pd.DataFrame) -> pd.DataFrame:
    """Compute features for a panel of prices."""
    features = pd.DataFrame(index=prices_df.index)
    for col in prices_df.columns:
        s = prices_df[col]
        features[f"{col}_return"] = s.diff()
        features[f"{col}_log_return"] = compute_log_returns(s)
        features[f"{col}_rv5"] = compute_realized_volatility(s.diff(), 5)
        features[f"{col}_roll_spread"] = compute_roll_spread(s)
    return features