wbt 0.3.0

pub mod core;

use std::collections::HashMap;
use std::io::Cursor;
use std::str::FromStr;

use polars::prelude::*;
use pyo3::exceptions::PyException;
use pyo3::prelude::*;
use pyo3::types::{PyBytes, PyBytesMethods, PyDict, PyList};
use serde_json::Value;

use crate::core::{WeightBacktest, WeightType};

// ---------------------------------------------------------------------------
// Arrow IPC <-> Polars DataFrame helpers
// ---------------------------------------------------------------------------

fn pyarrow_to_df(data: &[u8]) -> PyResult<DataFrame> {
    let cursor = Cursor::new(data);
    IpcReader::new(cursor)
        .finish()
        .map_err(|e| PyException::new_err(e.to_string()))
}

fn df_to_pyarrow(dataframe: &mut DataFrame) -> PyResult<Vec<u8>> {
    let mut buffer = Cursor::new(Vec::new());
    IpcWriter::new(&mut buffer)
        .finish(dataframe)
        .map_err(|e| PyException::new_err(e.to_string()))?;
    Ok(buffer.into_inner())
}

// ---------------------------------------------------------------------------
// HashMap<String, Value> -> PyDict helper
// ---------------------------------------------------------------------------

/// 将 `HashMap<String, Value>` 转成 `PyDict`，**按 key 字母顺序插入** 以保证 Python 端
/// 字典迭代顺序确定。值的转换委托给 [`value_to_py`]：
///
/// - `Value::Null` → Python `None`（注意：这是与历史 `_ => {}` 静默丢弃语义的区别；
///   现有 `*_stats` 调用方今天不会产出 `Value::Null`，但若未来某条路径用
///   `serde_json::Number::from_f64(NaN)` 等会得到 `Value::Null`，则在 Python 端
///   将变成 `None` 而非"缺失键"）。
/// - `Value::Bool` / `Value::Number` / `Value::String` 透传到对应 Python 原生类型。
/// - `Value::Array` 递归构造 `PyList`；`Value::Object` 递归构造 `PyDict`，
///   同样按 key 字母顺序插入以保持确定性。
fn hashmap_to_pydict<'py>(
    py: Python<'py>,
    map: &HashMap<String, Value>,
) -> PyResult<Bound<'py, PyDict>> {
    let dict = PyDict::new(py);
    let mut keys: Vec<&String> = map.keys().collect();
    keys.sort();
    for k in keys {
        dict.set_item(k, value_to_py(py, &map[k])?)?;
    }
    Ok(dict)
}

fn value_to_py<'py>(py: Python<'py>, v: &Value) -> PyResult<Bound<'py, pyo3::PyAny>> {
    use pyo3::IntoPyObject;
    Ok(match v {
        Value::Null => py.None().into_bound(py),
        Value::Bool(b) => b.into_pyobject(py)?.to_owned().into_any(),
        Value::Number(n) => {
            if let Some(i) = n.as_i64() {
                i.into_pyobject(py)?.into_any()
            } else if let Some(u) = n.as_u64() {
                u.into_pyobject(py)?.into_any()
            } else if let Some(f) = n.as_f64() {
                f.into_pyobject(py)?.into_any()
            } else {
                // 当前 serde_json 不会构造既非 i64 / u64 / f64 又有效的 Number；
                // 兜底为 None 以避免悄无声息地丢弃字段。
                py.None().into_bound(py)
            }
        }
        Value::String(s) => s.into_pyobject(py)?.into_any(),
        Value::Array(arr) => {
            let py_list = PyList::empty(py);
            for item in arr {
                py_list.append(value_to_py(py, item)?)?;
            }
            py_list.into_any()
        }
        Value::Object(obj) => {
            let py_dict = PyDict::new(py);
            let mut keys: Vec<&String> = obj.keys().collect();
            keys.sort();
            for key in keys {
                py_dict.set_item(key, value_to_py(py, &obj[key])?)?;
            }
            py_dict.into_any()
        }
    })
}

// ---------------------------------------------------------------------------
// PyWeightBacktest
// ---------------------------------------------------------------------------

#[pyclass(module = "wbt._wbt")]
#[repr(transparent)]
pub struct PyWeightBacktest {
    inner: WeightBacktest,
}

#[pymethods]
impl PyWeightBacktest {
    #[staticmethod]
    #[pyo3(signature = (data, digits=2, fee_rate=Some(0.0002), n_jobs=Some(4), weight_type="ts", yearly_days=252))]
    fn from_arrow<'py>(
        py: Python<'py>,
        data: Bound<'py, PyBytes>,
        digits: i64,
        fee_rate: Option<f64>,
        n_jobs: Option<usize>,
        weight_type: &str,
        yearly_days: usize,
    ) -> PyResult<Self> {
        let data = data.as_bytes();
        let df = pyarrow_to_df(data)?;
        let weight_type = WeightType::from_str(weight_type).unwrap_or(WeightType::TS);

        let mut inner = WeightBacktest::new(df, digits, fee_rate)
            .map_err(|e| PyException::new_err(e.to_string()))?;
        py.detach(|| {
            inner
                .backtest(n_jobs, weight_type, yearly_days)
                .map_err(|e| PyException::new_err(e.to_string()))
        })?;
        Ok(Self { inner })
    }

    fn stats<'py>(&self, py: Python<'py>) -> PyResult<Bound<'py, PyDict>> {
        let py_dict = PyDict::new(py);

        if let Some(ref report) = self.inner.report {
            let stats = &report.stats;

            let dp = &stats.daily_performance;
            let ep = &stats.evaluate_pairs;
            let pwr = &stats.period_win_rates;

            // 收益
            py_dict.set_item("绝对收益", dp.absolute_return)?;
            py_dict.set_item("年化收益", dp.annual_returns)?;
            py_dict.set_item("夏普比率", dp.sharpe_ratio)?;
            py_dict.set_item("卡玛比率", dp.calmar_ratio)?;
            py_dict.set_item("新高占比", dp.new_high_ratio)?;
            py_dict.set_item("单笔盈亏比", ep.single_profit_loss_ratio)?;
            py_dict.set_item("单笔收益", ep.single_trade_profit)?;
            py_dict.set_item("日胜率", dp.daily_win_rate)?;
            py_dict.set_item("周胜率", pwr.week)?;
            py_dict.set_item("月胜率", pwr.month)?;
            py_dict.set_item("季胜率", pwr.quarter)?;
            py_dict.set_item("年胜率", pwr.year)?;

            // 风险
            py_dict.set_item("最大回撤", dp.max_drawdown)?;
            py_dict.set_item("年化波动率", dp.annual_volatility)?;
            py_dict.set_item("下行波动率", dp.downside_volatility)?;
            py_dict.set_item("新高间隔", dp.new_high_interval)?;

            // 特质
            py_dict.set_item("交易次数", stats.trade_count)?;
            py_dict.set_item("年化交易次数", stats.annual_trade_count)?;
            py_dict.set_item("持仓K线数", ep.position_k_days)?;
            py_dict.set_item("交易胜率", ep.win_rate)?;
            py_dict.set_item("多头占比", stats.long_rate)?;
            py_dict.set_item("空头占比", stats.short_rate)?;
            py_dict.set_item("品种数量", stats.symbols_count)?;

            // 元数据
            py_dict.set_item("开始日期", stats.start_date.to_string())?;
            py_dict.set_item("结束日期", stats.end_date.to_string())?;
        }

        Ok(py_dict)
    }

    #[pyo3(text_signature = "($self)")]
    fn daily_return<'py>(&mut self, py: Python<'py>) -> PyResult<Bound<'py, PyBytes>> {
        let df = self
            .inner
            .daily_return_df()
            .map_err(|e| PyException::new_err(e.to_string()))?;
        let df_bytes = df_to_pyarrow(df)?;
        Ok(PyBytes::new(py, &df_bytes))
    }

    #[pyo3(signature = (min_days=120))]
    fn yearly_return<'py>(
        &mut self,
        py: Python<'py>,
        min_days: usize,
    ) -> PyResult<Bound<'py, PyBytes>> {
        let mut df = self
            .inner
            .yearly_return_df(min_days)
            .map_err(|e| PyException::new_err(e.to_string()))?;
        let df_bytes = df_to_pyarrow(&mut df)?;
        Ok(PyBytes::new(py, &df_bytes))
    }

    fn dailys<'py>(&mut self, py: Python<'py>) -> PyResult<Bound<'py, PyBytes>> {
        let df = self
            .inner
            .dailys_df()
            .map_err(|e| PyException::new_err(e.to_string()))?;
        let df_bytes = df_to_pyarrow(df)?;
        Ok(PyBytes::new(py, &df_bytes))
    }

    fn alpha<'py>(&mut self, py: Python<'py>) -> PyResult<Bound<'py, PyBytes>> {
        let mut df = self
            .inner
            .alpha_df()
            .map_err(|e| PyException::new_err(e.to_string()))?;
        let df_bytes = df_to_pyarrow(&mut df)?;
        Ok(PyBytes::new(py, &df_bytes))
    }

    #[pyo3(text_signature = "($self)")]
    fn pairs<'py>(&mut self, py: Python<'py>) -> PyResult<Bound<'py, PyBytes>> {
        match self
            .inner
            .pairs_df()
            .map_err(|e| PyException::new_err(e.to_string()))?
        {
            Some(df) => {
                let df_bytes = df_to_pyarrow(df)?;
                Ok(PyBytes::new(py, &df_bytes))
            }
            None => Ok(PyBytes::new(py, b"".as_slice())),
        }
    }

    /// 聚合去重后的开平记录表（Arrow IPC）。无 pairs 时返回空字节。
    #[pyo3(text_signature = "($self)")]
    fn aggregated_pairs<'py>(&mut self, py: Python<'py>) -> PyResult<Bound<'py, PyBytes>> {
        match self
            .inner
            .aggregated_pairs_df()
            .map_err(|e| PyException::new_err(e.to_string()))?
        {
            Some(mut df) => {
                let df_bytes = df_to_pyarrow(&mut df)?;
                Ok(PyBytes::new(py, &df_bytes))
            }
            None => Ok(PyBytes::new(py, b"".as_slice())),
        }
    }

    /// 每年最赚/最亏各 `top` 笔关键交易（Arrow IPC，含 year/kind 列）。无 pairs 时返回空字节。
    #[pyo3(signature = (top=3))]
    fn key_trades<'py>(&mut self, py: Python<'py>, top: usize) -> PyResult<Bound<'py, PyBytes>> {
        match self
            .inner
            .key_trades_df(top)
            .map_err(|e| PyException::new_err(e.to_string()))?
        {
            Some(mut df) => {
                let df_bytes = df_to_pyarrow(&mut df)?;
                Ok(PyBytes::new(py, &df_bytes))
            }
            None => Ok(PyBytes::new(py, b"".as_slice())),
        }
    }

    #[staticmethod]
    #[pyo3(signature = (path, digits=2, fee_rate=Some(0.0002), n_jobs=Some(4), weight_type="ts", yearly_days=252))]
    fn from_file<'py>(
        py: Python<'py>,
        path: &str,
        digits: i64,
        fee_rate: Option<f64>,
        n_jobs: Option<usize>,
        weight_type: &str,
        yearly_days: usize,
    ) -> PyResult<Self> {
        let weight_type_enum = WeightType::from_str(weight_type).unwrap_or(WeightType::TS);
        let mut inner = WeightBacktest::from_file(path, digits, fee_rate)
            .map_err(|e| PyException::new_err(e.to_string()))?;
        py.detach(|| {
            inner
                .backtest(n_jobs, weight_type_enum, yearly_days)
                .map_err(|e| PyException::new_err(e.to_string()))
        })?;
        Ok(Self { inner })
    }

    #[pyo3(text_signature = "($self)")]
    fn symbol_dict(&self) -> PyResult<Vec<String>> {
        if let Some(ref report) = self.inner.report {
            Ok(report.symbol_dict.clone())
        } else {
            Err(PyException::new_err("Report not found"))
        }
    }

    fn long_stats<'py>(&self, py: Python<'py>) -> PyResult<Bound<'py, PyDict>> {
        if let Some(ref report) = self.inner.report {
            hashmap_to_pydict(py, &report.long_stats)
        } else {
            Err(PyException::new_err("Report not found"))
        }
    }

    fn short_stats<'py>(&self, py: Python<'py>) -> PyResult<Bound<'py, PyDict>> {
        if let Some(ref report) = self.inner.report {
            hashmap_to_pydict(py, &report.short_stats)
        } else {
            Err(PyException::new_err("Report not found"))
        }
    }

    #[pyo3(signature = (sdt=None, edt=None, kind="多空"))]
    fn segment_stats<'py>(
        &self,
        py: Python<'py>,
        sdt: Option<i32>,
        edt: Option<i32>,
        kind: &str,
    ) -> PyResult<Bound<'py, PyDict>> {
        let map = self
            .inner
            .segment_stats(sdt, edt, kind)
            .map_err(|e| PyException::new_err(e.to_string()))?;
        hashmap_to_pydict(py, &map)
    }

    fn long_alpha_stats<'py>(&self, py: Python<'py>) -> PyResult<Bound<'py, PyDict>> {
        let map = self
            .inner
            .long_alpha_stats()
            .map_err(|e| PyException::new_err(e.to_string()))?;
        hashmap_to_pydict(py, &map)
    }

    #[pyo3(signature = (mode="history", target_vol=0.20, max_dd_threshold=0.20, min_year_days=120, recent_days=252, min_history_days=60))]
    #[allow(clippy::too_many_arguments)]
    fn is_good_strategy<'py>(
        &self,
        py: Python<'py>,
        mode: &str,
        target_vol: f64,
        max_dd_threshold: f64,
        min_year_days: usize,
        recent_days: usize,
        min_history_days: usize,
    ) -> PyResult<Bound<'py, PyDict>> {
        let map = self
            .inner
            .is_good_strategy(
                mode,
                target_vol,
                max_dd_threshold,
                min_year_days,
                recent_days,
                min_history_days,
            )
            .map_err(|e| PyException::new_err(e.to_string()))?;
        hashmap_to_pydict(py, &map)
    }
}

// ---------------------------------------------------------------------------
// daily_performance standalone function
// ---------------------------------------------------------------------------

/// 采用单利计算日收益数据的各项指标
#[pyfunction]
#[pyo3(signature = (daily_returns, yearly_days=None))]
pub fn daily_performance<'py>(
    py: Python<'py>,
    daily_returns: numpy::PyReadonlyArray1<'py, f64>,
    yearly_days: Option<usize>,
) -> PyResult<Bound<'py, PyDict>> {
    let daily_returns = daily_returns
        .as_slice()
        .map_err(|e| PyException::new_err(e.to_string()))?;
    let dp = crate::core::daily_performance::daily_performance(daily_returns, yearly_days)
        .map_err(|e| PyException::new_err(e.to_string()))?;

    let py_dict = PyDict::new(py);

    py_dict.set_item("绝对收益", dp.absolute_return)?;
    py_dict.set_item("年化", dp.annual_returns)?;
    py_dict.set_item("夏普", dp.sharpe_ratio)?;
    py_dict.set_item("最大回撤", dp.max_drawdown)?;
    py_dict.set_item("卡玛", dp.calmar_ratio)?;
    py_dict.set_item("日胜率", dp.daily_win_rate)?;
    py_dict.set_item("日盈亏比", dp.daily_profit_loss_ratio)?;
    py_dict.set_item("日赢面", dp.daily_win_probability)?;
    py_dict.set_item("年化波动率", dp.annual_volatility)?;
    py_dict.set_item("下行波动率", dp.downside_volatility)?;
    py_dict.set_item("非零覆盖", dp.non_zero_coverage)?;
    py_dict.set_item("盈亏平衡点", dp.break_even_point)?;
    py_dict.set_item("新高间隔", dp.new_high_interval)?;
    py_dict.set_item("新高占比", dp.new_high_ratio)?;
    py_dict.set_item("回撤风险", dp.drawdown_risk)?;
    py_dict.set_item("回归年度回报率", dp.annual_lin_reg_cumsum_return)?;
    py_dict.set_item(
        "长度调整平均最大回撤",
        dp.length_adjusted_average_max_drawdown,
    )?;

    Ok(py_dict)
}

// ---------------------------------------------------------------------------
// top_drawdowns standalone function
// ---------------------------------------------------------------------------

/// Identify the top-N drawdown windows in a return series. Input is
/// an Arrow IPC stream encoding a DataFrame with columns
/// ``date`` (Date or Datetime) and ``returns`` (f64); output is the
/// same encoding for the result DataFrame whose schema is
/// 回撤开始 / 回撤结束 / 回撤修复 / 净值回撤 / 回撤天数 / 恢复天数 / 新高间隔.
#[pyfunction]
#[pyo3(signature = (returns, top))]
pub fn top_drawdowns<'py>(
    py: Python<'py>,
    returns: Bound<'py, PyBytes>,
    top: usize,
) -> PyResult<Bound<'py, PyBytes>> {
    let data = returns.as_bytes();
    let df_in = pyarrow_to_df(data)?;

    let dt_col = df_in
        .column("date")
        .map_err(|e| PyException::new_err(format!("missing 'date' column: {e}")))?;
    let dt_type = dt_col.dtype();
    let dates: Vec<chrono::NaiveDate> = match dt_type {
        DataType::Datetime(_, _) => dt_col
            .datetime()
            .map_err(|e| PyException::new_err(e.to_string()))?
            .as_datetime_iter()
            .flatten()
            .map(|d| d.date())
            .collect(),
        DataType::Date => dt_col
            .date()
            .map_err(|e| PyException::new_err(e.to_string()))?
            .as_date_iter()
            .flatten()
            .collect(),
        _ => {
            return Err(PyException::new_err(format!(
                "Unsupported date dtype: {dt_type:?} (expected Date or Datetime)"
            )));
        }
    };

    let returns_vec: Vec<f64> = df_in
        .column("returns")
        .map_err(|e| PyException::new_err(format!("missing 'returns' column: {e}")))?
        .f64()
        .map_err(|e| PyException::new_err(e.to_string()))?
        .into_no_null_iter()
        .collect();

    let mut df_out = crate::core::top_drawdowns::top_drawdowns(&returns_vec, &dates, Some(top))
        .map_err(|e| PyException::new_err(e.to_string()))?;
    let bytes = df_to_pyarrow(&mut df_out)?;
    Ok(PyBytes::new(py, &bytes))
}

// ---------------------------------------------------------------------------
// cal_yearly_days standalone function
// ---------------------------------------------------------------------------

/// 计算年度交易日数量。输入为毫秒 unix 时间戳列表，返回年度交易日数。
#[pyfunction]
#[pyo3(signature = (timestamps_ms))]
pub fn cal_yearly_days(timestamps_ms: Vec<i64>) -> PyResult<i64> {
    use chrono::{DateTime, NaiveDate};
    if timestamps_ms.is_empty() {
        return Err(PyException::new_err("输入的日期数量必须大于0"));
    }
    let dates: Vec<NaiveDate> = timestamps_ms
        .iter()
        .filter_map(|ms| DateTime::from_timestamp_millis(*ms).map(|d| d.naive_utc().date()))
        .collect();
    let dropped = timestamps_ms.len() - dates.len();
    if dropped > 0 {
        log::warn!("cal_yearly_days: 丢弃了 {dropped} 个无法解析为日期的时间戳");
    }
    if dates.is_empty() {
        return Err(PyException::new_err("输入的日期数量必须大于0"));
    }
    Ok(crate::core::cal_yearly_days::cal_yearly_days(&dates))
}

// ---------------------------------------------------------------------------
// rolling_daily_performance standalone function
// ---------------------------------------------------------------------------

/// 计算滚动日收益的各项指标。输入为含 `dt` + `ret_col` 两列的 Arrow IPC 字节流，返回同格式 DataFrame。
#[pyfunction]
#[pyo3(signature = (data, ret_col, window=252, min_periods=100, yearly_days=None))]
pub fn rolling_daily_performance<'py>(
    py: Python<'py>,
    data: Bound<'py, PyBytes>,
    ret_col: &str,
    window: i64,
    min_periods: usize,
    yearly_days: Option<usize>,
) -> PyResult<Bound<'py, PyBytes>> {
    let df_in = pyarrow_to_df(data.as_bytes())?;

    let dt_col = df_in
        .column("dt")
        .map_err(|e| PyException::new_err(format!("missing 'dt' column: {e}")))?;
    if dt_col.null_count() > 0 {
        return Err(PyException::new_err(
            "'dt' column contains nulls; fill or drop them before calling rolling_daily_performance",
        ));
    }
    let dates: Vec<chrono::NaiveDate> = match dt_col.dtype() {
        DataType::Datetime(_, _) => dt_col
            .datetime()
            .map_err(|e| PyException::new_err(e.to_string()))?
            .as_datetime_iter()
            .flatten()
            .map(|d| d.date())
            .collect(),
        DataType::Date => dt_col
            .date()
            .map_err(|e| PyException::new_err(e.to_string()))?
            .as_date_iter()
            .flatten()
            .collect(),
        other => {
            return Err(PyException::new_err(format!(
                "Unsupported dt dtype: {other:?} (expected Date or Datetime)"
            )));
        }
    };

    // Null returns are converted to NaN here; the Rust core then maps NaN → 0,
    // matching czsc's `df[ret_col].fillna(0)` behavior. This intentionally diverges
    // from `top_drawdowns`, which rejects nulls via `into_no_null_iter`.
    let returns: Vec<f64> = df_in
        .column(ret_col)
        .map_err(|e| PyException::new_err(format!("missing '{ret_col}' column: {e}")))?
        .f64()
        .map_err(|e| PyException::new_err(e.to_string()))?
        .into_iter()
        .map(|opt| opt.unwrap_or(f64::NAN))
        .collect();

    let mut df_out = crate::core::rolling_daily_performance::rolling_daily_performance(
        dates,
        returns,
        window,
        min_periods,
        yearly_days,
    )
    .map_err(|e| PyException::new_err(e.to_string()))?;
    let bytes = df_to_pyarrow(&mut df_out)?;
    Ok(PyBytes::new(py, &bytes))
}

// ---------------------------------------------------------------------------
// Module registration
// ---------------------------------------------------------------------------

#[pymodule]
fn _wbt(m: &Bound<'_, PyModule>) -> PyResult<()> {
    // Bridge Rust log::warn! → Python logging (loguru 用户可一行接管)
    let _ = pyo3_log::try_init();

    m.add_class::<PyWeightBacktest>()?;
    m.add_function(wrap_pyfunction!(daily_performance, m)?)?;
    m.add_function(wrap_pyfunction!(top_drawdowns, m)?)?;
    m.add_function(wrap_pyfunction!(cal_yearly_days, m)?)?;
    m.add_function(wrap_pyfunction!(rolling_daily_performance, m)?)?;
    Ok(())
}