vector_ta/indicators/

percentile_nearest_rank.rs

#[cfg(feature = "python")]
use numpy::{IntoPyArray, PyArray1, PyArrayMethods, PyReadonlyArray1};
#[cfg(feature = "python")]
use pyo3::exceptions::PyValueError;
#[cfg(feature = "python")]
use pyo3::prelude::*;

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use serde::{Deserialize, Serialize};
#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use wasm_bindgen::prelude::*;

use crate::utilities::data_loader::{source_type, CandleFieldFlags, Candles};
use crate::utilities::enums::Kernel;
use crate::utilities::helpers::{
    alloc_with_nan_prefix, detect_best_batch_kernel, detect_best_kernel, init_matrix_prefixes,
    make_uninit_matrix,
};
#[cfg(feature = "python")]
use crate::utilities::kernel_validation::validate_kernel;

use std::convert::AsRef;
use std::error::Error;
use thiserror::Error;

#[cfg(not(target_arch = "wasm32"))]
use rayon::prelude::*;

impl<'a> AsRef<[f64]> for PercentileNearestRankInput<'a> {
    #[inline(always)]
    fn as_ref(&self) -> &[f64] {
        match &self.data {
            PercentileNearestRankData::Slice(slice) => slice,
            PercentileNearestRankData::Candles { candles, source } => source_type(candles, source),
        }
    }
}

#[derive(Debug, Clone)]
pub enum PercentileNearestRankData<'a> {
    Candles {
        candles: &'a Candles,
        source: &'a str,
    },
    Slice(&'a [f64]),
}

#[derive(Debug, Clone)]
pub struct PercentileNearestRankOutput {
    pub values: Vec<f64>,
}

#[derive(Debug, Clone)]
#[cfg_attr(
    all(target_arch = "wasm32", feature = "wasm"),
    derive(Serialize, Deserialize)
)]
pub struct PercentileNearestRankParams {
    pub length: Option<usize>,
    pub percentage: Option<f64>,
}

impl Default for PercentileNearestRankParams {
    fn default() -> Self {
        Self {
            length: Some(15),
            percentage: Some(50.0),
        }
    }
}

#[derive(Debug, Clone)]
pub struct PercentileNearestRankInput<'a> {
    pub data: PercentileNearestRankData<'a>,
    pub params: PercentileNearestRankParams,
}

impl<'a> PercentileNearestRankInput<'a> {
    #[inline]
    pub fn from_candles(c: &'a Candles, s: &'a str, p: PercentileNearestRankParams) -> Self {
        Self {
            data: PercentileNearestRankData::Candles {
                candles: c,
                source: s,
            },
            params: p,
        }
    }

    #[inline]
    pub fn from_slice(sl: &'a [f64], p: PercentileNearestRankParams) -> Self {
        Self {
            data: PercentileNearestRankData::Slice(sl),
            params: p,
        }
    }

    #[inline]
    pub fn with_default_candles(c: &'a Candles) -> Self {
        Self::from_candles(c, "close", PercentileNearestRankParams::default())
    }

    #[inline]
    pub fn get_length(&self) -> usize {
        self.params.length.unwrap_or(15)
    }

    #[inline]
    pub fn get_percentage(&self) -> f64 {
        self.params.percentage.unwrap_or(50.0)
    }
}

#[derive(Debug, Error)]
pub enum PercentileNearestRankError {
    #[error("percentile_nearest_rank: Input data is empty")]
    EmptyInputData,

    #[error("percentile_nearest_rank: All values are NaN")]
    AllValuesNaN,

    #[error(
        "percentile_nearest_rank: Invalid period: period = {period}, data length = {data_len}"
    )]
    InvalidPeriod { period: usize, data_len: usize },

    #[error("percentile_nearest_rank: Percentage must be between 0 and 100, got {percentage}")]
    InvalidPercentage { percentage: f64 },

    #[error("percentile_nearest_rank: Not enough valid data: needed = {needed}, valid = {valid}")]
    NotEnoughValidData { needed: usize, valid: usize },

    #[error("percentile_nearest_rank: Output length mismatch: expected {expected}, got {got}")]
    OutputLengthMismatch { expected: usize, got: usize },

    #[error("percentile_nearest_rank: Invalid range: start={start}, end={end}, step={step}")]
    InvalidRange {
        start: String,
        end: String,
        step: String,
    },

    #[error("percentile_nearest_rank: Invalid kernel for batch: {0:?}")]
    InvalidKernelForBatch(Kernel),
}

#[inline(always)]
fn pnr_prepare<'a>(
    input: &'a PercentileNearestRankInput,
    kernel: Kernel,
) -> Result<(&'a [f64], usize, f64, usize, Kernel), PercentileNearestRankError> {
    let data: &[f64] = input.as_ref();
    let len = data.len();
    if len == 0 {
        return Err(PercentileNearestRankError::EmptyInputData);
    }

    let first = data
        .iter()
        .position(|x| !x.is_nan())
        .ok_or(PercentileNearestRankError::AllValuesNaN)?;

    let length = input.get_length();
    let percentage = input.get_percentage();

    if length == 0 || length > len {
        return Err(PercentileNearestRankError::InvalidPeriod {
            period: length,
            data_len: len,
        });
    }
    if !(0.0..=100.0).contains(&percentage) || percentage.is_nan() || percentage.is_infinite() {
        return Err(PercentileNearestRankError::InvalidPercentage { percentage });
    }
    if len - first < length {
        return Err(PercentileNearestRankError::NotEnoughValidData {
            needed: length,
            valid: len - first,
        });
    }

    let chosen = match kernel {
        Kernel::Auto => Kernel::Scalar,
        k => k,
    };
    Ok((data, length, percentage, first, chosen))
}

#[inline(always)]
fn pnr_compute_into(
    data: &[f64],
    length: usize,
    percentage: f64,
    first: usize,
    _kernel: Kernel,
    out: &mut [f64],
) {
    let n = data.len();
    if n == 0 {
        return;
    }
    let start_i = first + length - 1;
    if start_i >= n {
        return;
    }

    let mut sorted: Vec<f64> = Vec::with_capacity(length);
    let window_start0 = start_i + 1 - length;
    for idx in window_start0..=start_i {
        let v = data[idx];
        if !v.is_nan() {
            sorted.push(v);
        }
    }
    sorted.sort_unstable_by(|a, b| a.partial_cmp(b).unwrap());

    let p_frac = percentage * 0.01;
    let k_const_usize = {
        let raw = (p_frac.mul_add(length as f64, 0.0)).round() as isize - 1;
        let mut k = if raw <= 0 { 0usize } else { raw as usize };
        if k >= length {
            k = length - 1;
        }
        k
    };
    let mut i = start_i;
    loop {
        if sorted.is_empty() {
            out[i] = f64::NAN;
        } else {
            let wl = sorted.len();
            let idx = if wl == length {
                k_const_usize
            } else {
                let raw = (p_frac.mul_add(wl as f64, 0.0)).round() as isize - 1;
                let mut k = if raw <= 0 { 0usize } else { raw as usize };
                if k >= wl {
                    k = wl - 1;
                }
                k
            };
            out[i] = sorted[idx];
        }

        if i + 1 >= n {
            break;
        }

        let out_idx = i + 1 - length;
        let v_out = data[out_idx];
        if !v_out.is_nan() {
            if let Ok(pos) = sorted.binary_search_by(|x| x.partial_cmp(&v_out).unwrap()) {
                sorted.remove(pos);
            }
        }
        let v_in = data[i + 1];
        if !v_in.is_nan() {
            match sorted.binary_search_by(|x| x.partial_cmp(&v_in).unwrap()) {
                Ok(pos) | Err(pos) => sorted.insert(pos, v_in),
            }
        }

        i += 1;
    }
}

#[inline]
pub fn percentile_nearest_rank(
    input: &PercentileNearestRankInput,
) -> Result<PercentileNearestRankOutput, PercentileNearestRankError> {
    percentile_nearest_rank_with_kernel(input, Kernel::Auto)
}

pub fn percentile_nearest_rank_with_kernel(
    input: &PercentileNearestRankInput,
    kernel: Kernel,
) -> Result<PercentileNearestRankOutput, PercentileNearestRankError> {
    let (data, length, percentage, first, chosen) = pnr_prepare(input, kernel)?;
    let warmup_end = first + length - 1;
    let mut out = alloc_with_nan_prefix(data.len(), warmup_end);
    pnr_compute_into(data, length, percentage, first, chosen, &mut out);
    Ok(PercentileNearestRankOutput { values: out })
}

#[cfg(not(all(target_arch = "wasm32", feature = "wasm")))]
#[inline]
pub fn percentile_nearest_rank_into(
    input: &PercentileNearestRankInput,
    out: &mut [f64],
) -> Result<(), PercentileNearestRankError> {
    percentile_nearest_rank_into_slice(out, input, Kernel::Auto)
}

#[inline]
pub fn percentile_nearest_rank_into_slice(
    dst: &mut [f64],
    input: &PercentileNearestRankInput,
    kernel: Kernel,
) -> Result<(), PercentileNearestRankError> {
    let (data, length, percentage, first, chosen) = pnr_prepare(input, kernel)?;
    if dst.len() != data.len() {
        return Err(PercentileNearestRankError::OutputLengthMismatch {
            expected: data.len(),
            got: dst.len(),
        });
    }

    pnr_compute_into(data, length, percentage, first, chosen, dst);

    let warmup_end = first + length - 1;
    for v in &mut dst[..warmup_end] {
        *v = f64::NAN;
    }
    Ok(())
}

#[derive(Copy, Clone, Debug)]
pub struct PercentileNearestRankBuilder {
    length: Option<usize>,
    percentage: Option<f64>,
    kernel: Kernel,
}

impl Default for PercentileNearestRankBuilder {
    fn default() -> Self {
        Self {
            length: None,
            percentage: None,
            kernel: Kernel::Auto,
        }
    }
}

impl PercentileNearestRankBuilder {
    #[inline(always)]
    pub fn new() -> Self {
        Self::default()
    }

    #[inline(always)]
    pub fn length(mut self, n: usize) -> Self {
        self.length = Some(n);
        self
    }

    #[inline(always)]
    pub fn percentage(mut self, p: f64) -> Self {
        self.percentage = Some(p);
        self
    }

    #[inline(always)]
    pub fn kernel(mut self, k: Kernel) -> Self {
        self.kernel = k;
        self
    }

    pub fn build(
        self,
        data: &[f64],
    ) -> Result<PercentileNearestRankOutput, PercentileNearestRankError> {
        let params = PercentileNearestRankParams {
            length: self.length,
            percentage: self.percentage,
        };
        let input = PercentileNearestRankInput::from_slice(data, params);
        percentile_nearest_rank_with_kernel(&input, self.kernel)
    }

    pub fn build_candles(
        self,
        candles: &Candles,
        source: &str,
    ) -> Result<PercentileNearestRankOutput, PercentileNearestRankError> {
        let params = PercentileNearestRankParams {
            length: self.length,
            percentage: self.percentage,
        };
        let input = PercentileNearestRankInput::from_candles(candles, source, params);
        percentile_nearest_rank_with_kernel(&input, self.kernel)
    }

    #[inline(always)]
    pub fn apply(
        self,
        c: &Candles,
    ) -> Result<PercentileNearestRankOutput, PercentileNearestRankError> {
        let p = PercentileNearestRankParams {
            length: self.length,
            percentage: self.percentage,
        };
        let i = PercentileNearestRankInput::from_candles(c, "close", p);
        percentile_nearest_rank_with_kernel(&i, self.kernel)
    }

    #[inline(always)]
    pub fn apply_slice(
        self,
        d: &[f64],
    ) -> Result<PercentileNearestRankOutput, PercentileNearestRankError> {
        let p = PercentileNearestRankParams {
            length: self.length,
            percentage: self.percentage,
        };
        let i = PercentileNearestRankInput::from_slice(d, p);
        percentile_nearest_rank_with_kernel(&i, self.kernel)
    }

    #[inline(always)]
    pub fn into_stream(self) -> Result<PercentileNearestRankStream, PercentileNearestRankError> {
        let p = PercentileNearestRankParams {
            length: self.length,
            percentage: self.percentage,
        };
        PercentileNearestRankStream::try_new(p)
    }

    pub fn with_default_slice(
        data: &[f64],
        k: Kernel,
    ) -> Result<PercentileNearestRankOutput, PercentileNearestRankError> {
        Self::new().kernel(k).apply_slice(data)
    }

    pub fn with_default_candles(
        c: &Candles,
    ) -> Result<PercentileNearestRankOutput, PercentileNearestRankError> {
        Self::new().kernel(Kernel::Auto).apply(c)
    }
}

use std::cmp::Reverse;
use std::collections::HashMap;

#[derive(Copy, Clone, Debug)]
struct FOrd(f64);
impl PartialEq for FOrd {
    #[inline]
    fn eq(&self, other: &Self) -> bool {
        self.0 == other.0
    }
}
impl Eq for FOrd {}
impl PartialOrd for FOrd {
    #[inline]
    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
        Some(self.cmp(other))
    }
}
impl Ord for FOrd {
    #[inline]
    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
        self.0.partial_cmp(&other.0).unwrap()
    }
}

#[derive(Copy, Clone, Debug, PartialEq, Eq, Hash)]
struct FKey(u64);
impl From<f64> for FKey {
    #[inline]
    fn from(x: f64) -> Self {
        let bits = if x == 0.0 { 0u64 } else { x.to_bits() };
        FKey(bits)
    }
}

#[derive(Debug, Clone)]
pub struct PercentileNearestRankStream {
    length: usize,
    percentage: f64,
    p_frac: f64,
    buffer: Vec<f64>,
    head: usize,
    filled: bool,

    left: std::collections::BinaryHeap<FOrd>,
    right: std::collections::BinaryHeap<Reverse<FOrd>>,
    delayed_left: HashMap<FKey, usize>,
    delayed_right: HashMap<FKey, usize>,
    size_left: usize,
    size_right: usize,

    t_full: usize,
}

#[inline(always)]
fn nearest_rank_index_fast(pf: f64, wl: usize) -> usize {
    let mut k = (pf.mul_add(wl as f64, 0.5)) as usize;
    if k == 0 {
        0
    } else {
        k -= 1;
        if k >= wl {
            wl - 1
        } else {
            k
        }
    }
}

impl PercentileNearestRankStream {
    pub fn try_new(
        params: PercentileNearestRankParams,
    ) -> Result<Self, PercentileNearestRankError> {
        let length = params.length.unwrap_or(15);
        if length == 0 {
            return Err(PercentileNearestRankError::InvalidPeriod {
                period: length,
                data_len: 0,
            });
        }
        let percentage = params.percentage.unwrap_or(50.0);
        if !(0.0..=100.0).contains(&percentage) || percentage.is_nan() || percentage.is_infinite() {
            return Err(PercentileNearestRankError::InvalidPercentage { percentage });
        }

        let p_frac = percentage * 0.01;
        let t_full = nearest_rank_index_fast(p_frac, length) + 1;

        Ok(Self {
            length,
            percentage,
            p_frac,
            buffer: vec![f64::NAN; length],
            head: 0,
            filled: false,
            left: std::collections::BinaryHeap::with_capacity(length),
            right: std::collections::BinaryHeap::with_capacity(length),
            delayed_left: HashMap::new(),
            delayed_right: HashMap::new(),
            size_left: 0,
            size_right: 0,
            t_full,
        })
    }

    #[inline(always)]
    fn prune_left(&mut self) {
        while let Some(&FOrd(x)) = self.left.peek() {
            let key = FKey::from(x);
            if let Some(cnt) = self.delayed_left.get_mut(&key) {
                if *cnt > 0 {
                    self.left.pop();
                    *cnt -= 1;
                    if *cnt == 0 {
                        self.delayed_left.remove(&key);
                    }
                } else {
                    break;
                }
            } else {
                break;
            }
        }
    }

    #[inline(always)]
    fn prune_right(&mut self) {
        while let Some(&Reverse(FOrd(x))) = self.right.peek() {
            let key = FKey::from(x);
            if let Some(cnt) = self.delayed_right.get_mut(&key) {
                if *cnt > 0 {
                    self.right.pop();
                    *cnt -= 1;
                    if *cnt == 0 {
                        self.delayed_right.remove(&key);
                    }
                } else {
                    break;
                }
            } else {
                break;
            }
        }
    }

    #[inline(always)]
    fn current_left_top(&mut self) -> Option<f64> {
        self.prune_left();
        self.left.peek().map(|v| v.0)
    }

    #[inline(always)]
    fn push_value(&mut self, v: f64) {
        if v.is_nan() {
            return;
        }
        if self.size_left == 0 {
            self.left.push(FOrd(v));
            self.size_left += 1;
        } else {
            let left_top = self.current_left_top().unwrap();
            if v <= left_top {
                self.left.push(FOrd(v));
                self.size_left += 1;
            } else {
                self.right.push(Reverse(FOrd(v)));
                self.size_right += 1;
            }
        }
    }

    #[inline(always)]
    fn erase_value(&mut self, v: f64) {
        if v.is_nan() {
            return;
        }
        let belongs_left = match self.current_left_top() {
            Some(top) => v <= top,
            None => false,
        };
        let key = FKey::from(v);
        if belongs_left {
            *self.delayed_left.entry(key).or_insert(0) += 1;
            if self.size_left > 0 {
                self.size_left -= 1;
            }
            self.prune_left();
        } else {
            *self.delayed_right.entry(key).or_insert(0) += 1;
            if self.size_right > 0 {
                self.size_right -= 1;
            }
            self.prune_right();
        }
    }

    #[inline(always)]
    fn target_left_for_valid(&self, valid: usize) -> usize {
        if valid == 0 {
            return 0;
        }
        if valid == self.length {
            return self.t_full;
        }
        nearest_rank_index_fast(self.p_frac, valid) + 1
    }

    #[inline(always)]
    fn rebalance(&mut self, target_left: usize) {
        self.prune_left();
        self.prune_right();

        while self.size_left > target_left {
            if let Some(FOrd(x)) = self.left.pop() {
                self.size_left -= 1;
                self.right.push(Reverse(FOrd(x)));
                self.size_right += 1;
            }
            self.prune_left();
        }
        while self.size_left < target_left {
            self.prune_right();
            if let Some(Reverse(FOrd(x))) = self.right.pop() {
                self.size_right -= 1;
                self.left.push(FOrd(x));
                self.size_left += 1;
            } else {
                break;
            }
        }
        self.prune_left();
    }

    pub fn update(&mut self, value: f64) -> Option<f64> {
        let outgoing = self.buffer[self.head];
        self.buffer[self.head] = value;
        self.head = (self.head + 1) % self.length;

        if !self.filled && self.head == 0 {
            self.filled = true;
        }

        self.push_value(value);

        if !self.filled {
            return None;
        }

        self.erase_value(outgoing);

        let valid = self.size_left + self.size_right;
        if valid == 0 {
            return Some(f64::NAN);
        }
        let target_left = self.target_left_for_valid(valid);
        self.rebalance(target_left);

        self.current_left_top().or(Some(f64::NAN))
    }
}

#[cfg(feature = "python")]
#[pyfunction(name = "percentile_nearest_rank")]
#[pyo3(signature = (data, length=15, percentage=50.0, kernel=None))]
pub fn percentile_nearest_rank_py<'py>(
    py: Python<'py>,
    data: PyReadonlyArray1<'py, f64>,
    length: usize,
    percentage: f64,
    kernel: Option<&str>,
) -> PyResult<Bound<'py, PyArray1<f64>>> {
    let kern = validate_kernel(kernel, false)?;
    let data_slice = data.as_slice()?;

    let params = PercentileNearestRankParams {
        length: Some(length),
        percentage: Some(percentage),
    };
    let input = PercentileNearestRankInput::from_slice(data_slice, params);

    let result = py
        .allow_threads(|| percentile_nearest_rank_with_kernel(&input, kern))
        .map_err(|e| PyValueError::new_err(e.to_string()))?;

    Ok(result.values.into_pyarray(py))
}

#[cfg(feature = "python")]
#[pyclass(name = "PercentileNearestRankStream")]
pub struct PercentileNearestRankStreamPy {
    stream: PercentileNearestRankStream,
}

#[cfg(feature = "python")]
#[pymethods]
impl PercentileNearestRankStreamPy {
    #[new]
    fn new(length: usize, percentage: f64) -> PyResult<Self> {
        let params = PercentileNearestRankParams {
            length: Some(length),
            percentage: Some(percentage),
        };
        let stream = PercentileNearestRankStream::try_new(params)
            .map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok(PercentileNearestRankStreamPy { stream })
    }

    fn update(&mut self, value: f64) -> Option<f64> {
        self.stream.update(value)
    }
}

#[cfg(feature = "python")]
#[pyfunction(name = "percentile_nearest_rank_batch")]
#[pyo3(signature = (data, length_range, percentage_range, kernel=None))]
pub fn percentile_nearest_rank_batch_py<'py>(
    py: Python<'py>,
    data: numpy::PyReadonlyArray1<'py, f64>,
    length_range: (usize, usize, usize),
    percentage_range: (f64, f64, f64),
    kernel: Option<&str>,
) -> PyResult<Bound<'py, pyo3::types::PyDict>> {
    use numpy::{IntoPyArray, PyArray1, PyArrayMethods};
    use pyo3::types::PyDict;

    let slice_in = data.as_slice()?;
    let sweep = PercentileNearestRankBatchRange {
        length: length_range,
        percentage: percentage_range,
    };

    let combos = expand_grid_pnr(&sweep).map_err(|e| PyValueError::new_err(e.to_string()))?;
    let rows = combos.len();
    let cols = slice_in.len();

    let total = rows
        .checked_mul(cols)
        .ok_or_else(|| PyValueError::new_err("rows*cols overflow"))?;
    let out_arr = unsafe { PyArray1::<f64>::new(py, [total], false) };
    let slice_out = unsafe { out_arr.as_slice_mut()? };

    for (row_idx, combo) in combos.iter().enumerate() {
        let length = combo.length.unwrap_or(15);
        let warmup = length - 1;
        let row_start = row_idx * cols;
        for i in 0..warmup.min(cols) {
            slice_out[row_start + i] = f64::NAN;
        }
    }

    let kern = validate_kernel(kernel, true)?;
    py.allow_threads(|| {
        let k = match kern {
            Kernel::Auto => detect_best_batch_kernel(),
            k => k,
        };

        pnr_batch_inner_into(slice_in, &combos, k, true, slice_out)
    })
    .map_err(|e| PyValueError::new_err(e.to_string()))?;

    let dict = PyDict::new(py);
    dict.set_item("values", out_arr.reshape((rows, cols))?)?;
    dict.set_item(
        "lengths",
        combos
            .iter()
            .map(|p| p.length.unwrap_or(15) as u64)
            .collect::<Vec<_>>()
            .into_pyarray(py),
    )?;
    dict.set_item(
        "percentages",
        combos
            .iter()
            .map(|p| p.percentage.unwrap_or(50.0))
            .collect::<Vec<_>>()
            .into_pyarray(py),
    )?;
    Ok(dict.into())
}

#[derive(Clone, Debug)]
pub struct PercentileNearestRankBatchRange {
    pub length: (usize, usize, usize),
    pub percentage: (f64, f64, f64),
}

impl Default for PercentileNearestRankBatchRange {
    fn default() -> Self {
        Self {
            length: (15, 264, 1),
            percentage: (50.0, 50.0, 0.0),
        }
    }
}

#[derive(Clone, Debug, Default)]
pub struct PercentileNearestRankBatchBuilder {
    range: PercentileNearestRankBatchRange,
    kernel: Kernel,
}

impl PercentileNearestRankBatchBuilder {
    pub fn new() -> Self {
        Self::default()
    }

    pub fn kernel(mut self, k: Kernel) -> Self {
        self.kernel = k;
        self
    }

    pub fn period_range(mut self, start: usize, end: usize, step: usize) -> Self {
        self.range.length = (start, end, step);
        self
    }

    pub fn percentage_range(mut self, start: f64, end: f64, step: f64) -> Self {
        self.range.percentage = (start, end, step);
        self
    }

    pub fn apply(
        self,
        data: &[f64],
    ) -> Result<PercentileNearestRankBatchOutput, PercentileNearestRankError> {
        pnr_batch_slice(data, &self.range, self.kernel)
    }

    pub fn apply_candles(
        self,
        candles: &Candles,
        source: &str,
    ) -> Result<PercentileNearestRankBatchOutput, PercentileNearestRankError> {
        let data = source_type(candles, source);
        pnr_batch_slice(data, &self.range, self.kernel)
    }

    pub fn with_default_slice(
        data: &[f64],
        k: Kernel,
    ) -> Result<PercentileNearestRankBatchOutput, PercentileNearestRankError> {
        PercentileNearestRankBatchBuilder::new()
            .kernel(k)
            .apply(data)
    }

    pub fn with_default_candles(
        c: &Candles,
    ) -> Result<PercentileNearestRankBatchOutput, PercentileNearestRankError> {
        PercentileNearestRankBatchBuilder::new()
            .kernel(Kernel::Auto)
            .apply_candles(c, "close")
    }
}

#[derive(Clone, Debug)]
pub struct PercentileNearestRankBatchOutput {
    pub values: Vec<f64>,
    pub combos: Vec<PercentileNearestRankParams>,
    pub rows: usize,
    pub cols: usize,
}

impl PercentileNearestRankBatchOutput {
    pub fn row_for_params(&self, p: &PercentileNearestRankParams) -> Option<usize> {
        self.combos.iter().position(|c| {
            c.length.unwrap_or(15) == p.length.unwrap_or(15)
                && (c.percentage.unwrap_or(50.0) - p.percentage.unwrap_or(50.0)).abs() < 1e-12
        })
    }

    pub fn values_for(&self, p: &PercentileNearestRankParams) -> Option<&[f64]> {
        self.row_for_params(p).map(|row| {
            let start = row * self.cols;
            &self.values[start..start + self.cols]
        })
    }
}

#[inline(always)]
fn expand_grid_pnr(
    r: &PercentileNearestRankBatchRange,
) -> Result<Vec<PercentileNearestRankParams>, PercentileNearestRankError> {
    fn axis_usize(
        (start, end, step): (usize, usize, usize),
    ) -> Result<Vec<usize>, PercentileNearestRankError> {
        if step == 0 || start == end {
            return Ok(vec![start]);
        }
        if start < end {
            return Ok((start..=end).step_by(step.max(1)).collect());
        }

        let mut v = Vec::new();
        let mut x = start as isize;
        let end_i = end as isize;
        let st = (step as isize).max(1);
        while x >= end_i {
            v.push(x as usize);
            x -= st;
        }
        if v.is_empty() {
            return Err(PercentileNearestRankError::InvalidRange {
                start: start.to_string(),
                end: end.to_string(),
                step: step.to_string(),
            });
        }
        Ok(v)
    }

    fn axis_f64(
        (start, end, step): (f64, f64, f64),
    ) -> Result<Vec<f64>, PercentileNearestRankError> {
        if step.abs() < 1e-12 || (start - end).abs() < 1e-12 {
            return Ok(vec![start]);
        }
        if start < end {
            let mut v = Vec::new();
            let mut x = start;
            let st = step.abs();
            while x <= end + 1e-12 {
                v.push(x);
                x += st;
            }
            if v.is_empty() {
                return Err(PercentileNearestRankError::InvalidRange {
                    start: start.to_string(),
                    end: end.to_string(),
                    step: step.to_string(),
                });
            }
            return Ok(v);
        }
        let mut v = Vec::new();
        let mut x = start;
        let st = step.abs();
        while x + 1e-12 >= end {
            v.push(x);
            x -= st;
        }
        if v.is_empty() {
            return Err(PercentileNearestRankError::InvalidRange {
                start: start.to_string(),
                end: end.to_string(),
                step: step.to_string(),
            });
        }
        Ok(v)
    }

    let lengths = axis_usize(r.length)?;
    let percentages = axis_f64(r.percentage)?;

    let cap = lengths
        .len()
        .checked_mul(percentages.len())
        .ok_or_else(|| PercentileNearestRankError::InvalidRange {
            start: "cap".into(),
            end: "overflow".into(),
            step: "mul".into(),
        })?;

    if cap == 0 {
        return Err(PercentileNearestRankError::InvalidRange {
            start: "range".into(),
            end: "range".into(),
            step: "empty".into(),
        });
    }

    let mut combos = Vec::with_capacity(cap);
    for &length in &lengths {
        for &percentage in &percentages {
            combos.push(PercentileNearestRankParams {
                length: Some(length),
                percentage: Some(percentage),
            });
        }
    }
    Ok(combos)
}

#[inline(always)]
pub fn pnr_batch_with_kernel(
    data: &[f64],
    sweep: &PercentileNearestRankBatchRange,
    k: Kernel,
) -> Result<PercentileNearestRankBatchOutput, PercentileNearestRankError> {
    let kernel = match k {
        Kernel::Auto => detect_best_batch_kernel(),
        other if other.is_batch() => other,
        _ => return Err(PercentileNearestRankError::InvalidKernelForBatch(k)),
    };
    pnr_batch_inner(data, sweep, kernel, true)
}

#[inline(always)]
pub fn pnr_batch_slice(
    data: &[f64],
    sweep: &PercentileNearestRankBatchRange,
    k: Kernel,
) -> Result<PercentileNearestRankBatchOutput, PercentileNearestRankError> {
    pnr_batch_inner(data, sweep, k, false)
}

#[inline(always)]
pub fn pnr_batch_par_slice(
    data: &[f64],
    sweep: &PercentileNearestRankBatchRange,
    k: Kernel,
) -> Result<PercentileNearestRankBatchOutput, PercentileNearestRankError> {
    pnr_batch_inner(data, sweep, k, true)
}

#[inline(always)]
fn pnr_batch_inner(
    data: &[f64],
    sweep: &PercentileNearestRankBatchRange,
    kern: Kernel,
    parallel: bool,
) -> Result<PercentileNearestRankBatchOutput, PercentileNearestRankError> {
    if data.is_empty() {
        return Err(PercentileNearestRankError::EmptyInputData);
    }
    let combos = expand_grid_pnr(sweep)?;
    if combos.is_empty() {
        return Err(PercentileNearestRankError::InvalidRange {
            start: "range".into(),
            end: "range".into(),
            step: "empty".into(),
        });
    }
    let rows = combos.len();
    let cols = data.len();

    let first = data
        .iter()
        .position(|x| !x.is_nan())
        .ok_or(PercentileNearestRankError::AllValuesNaN)?;
    let max_len = combos.iter().map(|c| c.length.unwrap()).max().unwrap();
    if data.len() - first < max_len {
        return Err(PercentileNearestRankError::NotEnoughValidData {
            needed: max_len,
            valid: data.len() - first,
        });
    }

    let _ = rows
        .checked_mul(cols)
        .ok_or_else(|| PercentileNearestRankError::InvalidRange {
            start: rows.to_string(),
            end: cols.to_string(),
            step: "rows*cols".into(),
        })?;

    let mut buf_mu = make_uninit_matrix(rows, cols);
    let warm: Vec<usize> = combos
        .iter()
        .map(|c| first + c.length.unwrap() - 1)
        .collect();
    init_matrix_prefixes(&mut buf_mu, cols, &warm);

    let mut guard = core::mem::ManuallyDrop::new(buf_mu);
    let out: &mut [f64] =
        unsafe { core::slice::from_raw_parts_mut(guard.as_mut_ptr() as *mut f64, guard.len()) };

    pnr_batch_inner_into(data, &combos, kern, parallel, out)?;

    let values = unsafe {
        Vec::from_raw_parts(
            guard.as_mut_ptr() as *mut f64,
            guard.len(),
            guard.capacity(),
        )
    };

    Ok(PercentileNearestRankBatchOutput {
        values,
        combos,
        rows,
        cols,
    })
}

#[inline(always)]
fn pnr_batch_inner_into(
    data: &[f64],
    combos: &[PercentileNearestRankParams],
    kern: Kernel,
    parallel: bool,
    out: &mut [f64],
) -> Result<(), PercentileNearestRankError> {
    let cols = data.len();
    if cols == 0 {
        return Err(PercentileNearestRankError::EmptyInputData);
    }

    let first = data
        .iter()
        .position(|x| !x.is_nan())
        .ok_or(PercentileNearestRankError::AllValuesNaN)?;
    let chosen = match kern {
        Kernel::Auto => Kernel::Scalar,
        k => k,
    };

    use std::collections::HashMap;
    let mut by_len: HashMap<usize, Vec<(usize, f64)>> = HashMap::new();
    for (row, p) in combos.iter().enumerate() {
        let len = p.length.unwrap_or(15);
        let perc = p.percentage.unwrap_or(50.0);
        by_len.entry(len).or_default().push((row, perc));
    }

    let has_benefit = by_len.values().any(|v| v.len() > 1);

    if parallel || !has_benefit {
        let do_row = |row: usize, dst_row: &mut [f64]| {
            let length = combos[row].length.unwrap_or(15);
            let percentage = combos[row].percentage.unwrap_or(50.0);

            pnr_compute_into(data, length, percentage, first, chosen, dst_row);
        };

        if parallel {
            #[cfg(not(target_arch = "wasm32"))]
            {
                use rayon::prelude::*;
                out.par_chunks_mut(cols)
                    .enumerate()
                    .for_each(|(row, s)| do_row(row, s));
            }
            #[cfg(target_arch = "wasm32")]
            for (row, s) in out.chunks_mut(cols).enumerate() {
                do_row(row, s);
            }
        } else {
            for (row, s) in out.chunks_mut(cols).enumerate() {
                do_row(row, s);
            }
        }
    } else {
        for (length, rows) in by_len.into_iter() {
            let start_i = first + length - 1;
            if start_i >= cols {
                continue;
            }

            let mut rows_info: Vec<(usize, f64, usize)> = Vec::with_capacity(rows.len());
            for &(row, perc) in &rows {
                let p_frac = perc * 0.01;
                let raw = (p_frac.mul_add(length as f64, 0.0)).round() as isize - 1;
                let mut k = if raw <= 0 { 0usize } else { raw as usize };
                if k >= length {
                    k = length - 1;
                }
                rows_info.push((row, p_frac, k));
            }

            let mut sorted: Vec<f64> = Vec::with_capacity(length);
            let window_start0 = start_i + 1 - length;
            for idx in window_start0..=start_i {
                let v = data[idx];
                if !v.is_nan() {
                    sorted.push(v);
                }
            }
            sorted.sort_unstable_by(|a, b| a.partial_cmp(b).unwrap());

            let mut i = start_i;
            loop {
                if sorted.is_empty() {
                    for &(row, _, _) in &rows_info {
                        out[row * cols + i] = f64::NAN;
                    }
                } else {
                    let wl = sorted.len();
                    let full = wl == length;
                    for &(row, p_frac, k_const) in &rows_info {
                        let idx = if full {
                            k_const
                        } else {
                            let raw = (p_frac.mul_add(wl as f64, 0.0)).round() as isize - 1;
                            let mut k = if raw <= 0 { 0usize } else { raw as usize };
                            if k >= wl {
                                k = wl - 1;
                            }
                            k
                        };
                        out[row * cols + i] = sorted[idx];
                    }
                }

                if i + 1 >= cols {
                    break;
                }

                let out_idx = i + 1 - length;
                let v_out = data[out_idx];
                if !v_out.is_nan() {
                    if let Ok(pos) = sorted.binary_search_by(|x| x.partial_cmp(&v_out).unwrap()) {
                        sorted.remove(pos);
                    }
                }
                let v_in = data[i + 1];
                if !v_in.is_nan() {
                    match sorted.binary_search_by(|x| x.partial_cmp(&v_in).unwrap()) {
                        Ok(pos) | Err(pos) => sorted.insert(pos, v_in),
                    }
                }

                i += 1;
            }
        }
    }
    Ok(())
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn percentile_nearest_rank_js(
    data: &[f64],
    length: usize,
    percentage: f64,
) -> Result<Vec<f64>, JsValue> {
    let params = PercentileNearestRankParams {
        length: Some(length),
        percentage: Some(percentage),
    };
    let input = PercentileNearestRankInput::from_slice(data, params);
    percentile_nearest_rank(&input)
        .map(|o| o.values)
        .map_err(|e| JsValue::from_str(&e.to_string()))
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn percentile_nearest_rank_alloc(n: usize) -> *mut f64 {
    let mut v = Vec::<f64>::with_capacity(n);
    let p = v.as_mut_ptr();
    core::mem::forget(v);
    p
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn percentile_nearest_rank_free(ptr: *mut f64, n: usize) {
    unsafe {
        let _ = Vec::from_raw_parts(ptr, n, n);
    }
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn percentile_nearest_rank_into(
    data_ptr: *const f64,
    out_ptr: *mut f64,
    len: usize,
    length: usize,
    percentage: f64,
) -> Result<(), JsValue> {
    if data_ptr.is_null() || out_ptr.is_null() {
        return Err(JsValue::from_str("null pointer"));
    }
    unsafe {
        let data = std::slice::from_raw_parts(data_ptr, len);
        let params = PercentileNearestRankParams {
            length: Some(length),
            percentage: Some(percentage),
        };
        let input = PercentileNearestRankInput::from_slice(data, params);

        if data_ptr == out_ptr {
            let mut temp = vec![0.0; len];
            percentile_nearest_rank_into_slice(&mut temp, &input, detect_best_kernel())
                .map_err(|e| JsValue::from_str(&e.to_string()))?;
            let out = std::slice::from_raw_parts_mut(out_ptr, len);
            out.copy_from_slice(&temp);
        } else {
            let out = std::slice::from_raw_parts_mut(out_ptr, len);
            percentile_nearest_rank_into_slice(out, &input, detect_best_kernel())
                .map_err(|e| JsValue::from_str(&e.to_string()))?;
        }
        Ok(())
    }
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[derive(Serialize, Deserialize)]
pub struct PercentileNearestRankBatchConfig {
    pub length_range: (usize, usize, usize),
    pub percentage_range: (f64, f64, f64),
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[derive(Serialize, Deserialize)]
pub struct PercentileNearestRankBatchJsOutput {
    pub values: Vec<f64>,
    pub combos: Vec<PercentileNearestRankParams>,
    pub rows: usize,
    pub cols: usize,
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen(js_name = percentile_nearest_rank_batch)]
pub fn percentile_nearest_rank_batch_unified_js(
    data: &[f64],
    config: JsValue,
) -> Result<JsValue, JsValue> {
    let cfg: PercentileNearestRankBatchConfig = serde_wasm_bindgen::from_value(config)
        .map_err(|e| JsValue::from_str(&format!("Invalid config: {}", e)))?;
    let sweep = PercentileNearestRankBatchRange {
        length: cfg.length_range,
        percentage: cfg.percentage_range,
    };
    let out = pnr_batch_inner(data, &sweep, detect_best_batch_kernel(), false)
        .map_err(|e| JsValue::from_str(&e.to_string()))?;
    let js = PercentileNearestRankBatchJsOutput {
        values: out.values,
        combos: out.combos,
        rows: out.rows,
        cols: out.cols,
    };
    serde_wasm_bindgen::to_value(&js)
        .map_err(|e| JsValue::from_str(&format!("Serialization error: {}", e)))
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::utilities::data_loader::read_candles_from_csv;
    use std::error::Error;

    macro_rules! skip_if_unsupported {
        ($kernel:expr, $test_name:expr) => {
            #[cfg(not(all(feature = "nightly-avx", target_arch = "x86_64")))]
            match $kernel {
                Kernel::Avx2 | Kernel::Avx512 | Kernel::Avx2Batch | Kernel::Avx512Batch => {
                    println!("[{}] Skipping: AVX not supported", $test_name);
                    return Ok(());
                }
                _ => {}
            }
        };
    }

    fn check_pnr_accuracy(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;

        let params = PercentileNearestRankParams {
            length: Some(15),
            percentage: Some(50.0),
        };
        let input = PercentileNearestRankInput::from_candles(&candles, "close", params);
        let result = percentile_nearest_rank_with_kernel(&input, kernel)?;

        assert_eq!(result.values.len(), candles.close.len());

        for i in 0..14 {
            assert!(
                result.values[i].is_nan(),
                "[{}] Expected NaN at index {}",
                test_name,
                i
            );
        }

        assert!(
            !result.values[14].is_nan(),
            "[{}] Expected valid value at index 14",
            test_name
        );

        let expected_last_5 = vec![59419.0, 59419.0, 59300.0, 59285.0, 59273.0];
        let len = result.values.len();
        let actual_last_5 = &result.values[len - 5..];

        for (i, (&actual, &expected)) in
            actual_last_5.iter().zip(expected_last_5.iter()).enumerate()
        {
            let diff = (actual - expected).abs();
            assert!(
                diff < 1e-6,
                "[{}] Value mismatch at last_5[{}]: expected {}, got {}, diff {}",
                test_name,
                i,
                expected,
                actual,
                diff
            );
        }

        Ok(())
    }

    fn check_pnr_partial_params(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let data = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0];

        let params = PercentileNearestRankParams {
            length: Some(5),
            percentage: None,
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank_with_kernel(&input, kernel)?;

        assert_eq!(result.values.len(), data.len());
        assert_eq!(result.values[4], 3.0);

        Ok(())
    }

    fn check_pnr_default_candles(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;

        let input = PercentileNearestRankInput::with_default_candles(&candles);
        let result = percentile_nearest_rank_with_kernel(&input, kernel)?;

        assert_eq!(result.values.len(), candles.close.len());
        Ok(())
    }

    fn check_pnr_zero_period(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let data = vec![1.0; 10];
        let params = PercentileNearestRankParams {
            length: Some(0),
            percentage: Some(50.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank_with_kernel(&input, kernel);

        assert!(matches!(
            result,
            Err(PercentileNearestRankError::InvalidPeriod { .. })
        ));
        Ok(())
    }

    fn check_pnr_period_exceeds_length(
        test_name: &str,
        kernel: Kernel,
    ) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let data = vec![1.0; 5];
        let params = PercentileNearestRankParams {
            length: Some(10),
            percentage: Some(50.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank_with_kernel(&input, kernel);

        assert!(matches!(
            result,
            Err(PercentileNearestRankError::InvalidPeriod { .. })
        ));
        Ok(())
    }

    fn check_pnr_very_small_dataset(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let data = vec![5.0];
        let params = PercentileNearestRankParams {
            length: Some(1),
            percentage: Some(50.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank_with_kernel(&input, kernel)?;

        assert_eq!(result.values.len(), 1);
        assert_eq!(result.values[0], 5.0);
        Ok(())
    }

    fn check_pnr_empty_input(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let data: Vec<f64> = vec![];
        let params = PercentileNearestRankParams::default();
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank_with_kernel(&input, kernel);

        assert!(matches!(
            result,
            Err(PercentileNearestRankError::EmptyInputData)
        ));
        Ok(())
    }

    fn check_pnr_invalid_percentage(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let data = vec![1.0; 20];

        let params = PercentileNearestRankParams {
            length: Some(5),
            percentage: Some(150.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank_with_kernel(&input, kernel);
        assert!(matches!(
            result,
            Err(PercentileNearestRankError::InvalidPercentage { .. })
        ));

        let params = PercentileNearestRankParams {
            length: Some(5),
            percentage: Some(-10.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank_with_kernel(&input, kernel);
        assert!(matches!(
            result,
            Err(PercentileNearestRankError::InvalidPercentage { .. })
        ));

        Ok(())
    }

    fn check_pnr_reinput(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;

        let first_params = PercentileNearestRankParams {
            length: Some(15),
            percentage: Some(50.0),
        };
        let first_input = PercentileNearestRankInput::from_candles(&candles, "close", first_params);
        let first_result = percentile_nearest_rank_with_kernel(&first_input, kernel)?;

        let second_params = PercentileNearestRankParams {
            length: Some(15),
            percentage: Some(50.0),
        };
        let second_input =
            PercentileNearestRankInput::from_slice(&first_result.values, second_params);
        let second_result = percentile_nearest_rank_with_kernel(&second_input, kernel)?;

        assert_eq!(second_result.values.len(), first_result.values.len());
        Ok(())
    }

    fn check_pnr_nan_handling(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let data = vec![
            1.0,
            2.0,
            f64::NAN,
            4.0,
            5.0,
            f64::NAN,
            7.0,
            8.0,
            9.0,
            10.0,
            11.0,
            12.0,
            13.0,
            f64::NAN,
            15.0,
        ];

        let params = PercentileNearestRankParams {
            length: Some(5),
            percentage: Some(50.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank_with_kernel(&input, kernel)?;

        assert_eq!(result.values.len(), data.len());

        assert!(!result.values[6].is_nan());
        Ok(())
    }

    fn check_pnr_streaming(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;

        let params = PercentileNearestRankParams {
            length: Some(15),
            percentage: Some(50.0),
        };

        let input = PercentileNearestRankInput::from_candles(&candles, "close", params.clone());
        let batch_output = percentile_nearest_rank_with_kernel(&input, kernel)?.values;

        let mut stream = PercentileNearestRankStream::try_new(params)?;

        let mut stream_values = Vec::with_capacity(candles.close.len());
        for &price in &candles.close {
            match stream.update(price) {
                Some(val) => stream_values.push(val),
                None => stream_values.push(f64::NAN),
            }
        }

        assert_eq!(batch_output.len(), stream_values.len());
        for (i, (&b, &s)) in batch_output.iter().zip(stream_values.iter()).enumerate() {
            if b.is_nan() && s.is_nan() {
                continue;
            }
            let diff = (b - s).abs();
            assert!(
                diff < 1e-9,
                "[{}] PNR streaming mismatch at idx {}: batch={}, stream={}, diff={}",
                test_name,
                i,
                b,
                s,
                diff
            );
        }
        Ok(())
    }

    #[cfg(debug_assertions)]
    fn check_pnr_no_poison(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;

        let test_params = vec![
            PercentileNearestRankParams::default(),
            PercentileNearestRankParams {
                length: Some(5),
                percentage: Some(25.0),
            },
            PercentileNearestRankParams {
                length: Some(10),
                percentage: Some(75.0),
            },
            PercentileNearestRankParams {
                length: Some(20),
                percentage: Some(50.0),
            },
            PercentileNearestRankParams {
                length: Some(50),
                percentage: Some(90.0),
            },
        ];

        for (param_idx, params) in test_params.iter().enumerate() {
            let input = PercentileNearestRankInput::from_candles(&candles, "close", params.clone());
            let output = percentile_nearest_rank_with_kernel(&input, kernel)?;

            for (i, &val) in output.values.iter().enumerate() {
                if val.is_nan() {
                    continue;
                }

                let bits = val.to_bits();

                if bits == 0x11111111_11111111 {
                    panic!(
                        "[{}] Found alloc_with_nan_prefix poison value {} (0x{:016X}) at index {} \
                        with params: length={}, percentage={}",
                        test_name,
                        val,
                        bits,
                        i,
                        params.length.unwrap_or(15),
                        params.percentage.unwrap_or(50.0)
                    );
                }
            }
        }

        Ok(())
    }

    #[cfg(not(debug_assertions))]
    fn check_pnr_no_poison(_test_name: &str, _kernel: Kernel) -> Result<(), Box<dyn Error>> {
        Ok(())
    }

    macro_rules! generate_all_pnr_tests {
        ($($test_fn:ident),*) => {
            paste::paste! {
                $(
                    #[test]
                    fn [<$test_fn _scalar_f64>]() {
                        let _ = $test_fn(stringify!([<$test_fn _scalar_f64>]), Kernel::Scalar);
                    }
                )*
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                $(
                    #[test]
                    fn [<$test_fn _avx2_f64>]() {
                        let _ = $test_fn(stringify!([<$test_fn _avx2_f64>]), Kernel::Avx2);
                    }
                    #[test]
                    fn [<$test_fn _avx512_f64>]() {
                        let _ = $test_fn(stringify!([<$test_fn _avx512_f64>]), Kernel::Avx512);
                    }
                )*
                #[cfg(all(target_arch = "wasm32", target_feature = "simd128"))]
                $(
                    #[test]
                    fn [<$test_fn _simd128_f64>]() {
                        let _ = $test_fn(stringify!([<$test_fn _simd128_f64>]), Kernel::Scalar);
                    }
                )*
            }
        }
    }

    generate_all_pnr_tests!(
        check_pnr_accuracy,
        check_pnr_partial_params,
        check_pnr_default_candles,
        check_pnr_zero_period,
        check_pnr_period_exceeds_length,
        check_pnr_very_small_dataset,
        check_pnr_empty_input,
        check_pnr_invalid_percentage,
        check_pnr_reinput,
        check_pnr_nan_handling,
        check_pnr_streaming,
        check_pnr_no_poison
    );

    #[test]
    fn test_percentile_nearest_rank_into_matches_api() {
        let mut data = Vec::with_capacity(256);
        data.extend_from_slice(&[f64::NAN, f64::NAN, f64::NAN]);
        for i in 0..253 {
            data.push((i as f64) * 0.5 + ((i % 7) as f64) * 0.1);
        }

        let params = PercentileNearestRankParams {
            length: Some(15),
            percentage: Some(50.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);

        let base = percentile_nearest_rank(&input).expect("baseline ok").values;

        let mut out = vec![0.0; data.len()];
        let _ = percentile_nearest_rank_into(&input, &mut out).expect("into ok");

        assert_eq!(base.len(), out.len());
        for (i, (&a, &b)) in base.iter().zip(out.iter()).enumerate() {
            let eq = (a.is_nan() && b.is_nan()) || (a == b);
            assert!(eq, "mismatch at {}: base={} out={}", i, a, b);
        }
    }

    fn check_batch_default_row(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let file = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let c = read_candles_from_csv(file)?;

        let output = PercentileNearestRankBatchBuilder::new()
            .kernel(kernel)
            .apply_candles(&c, "close")?;

        let def = PercentileNearestRankParams::default();
        let row = output.values_for(&def).expect("default row missing");

        assert_eq!(row.len(), c.close.len());

        for i in 0..14 {
            assert!(row[i].is_nan());
        }
        assert!(!row[14].is_nan());

        Ok(())
    }

    fn check_batch_sweep(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let file = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let c = read_candles_from_csv(file)?;

        let output = PercentileNearestRankBatchBuilder::new()
            .kernel(kernel)
            .period_range(10, 30, 10)
            .percentage_range(25.0, 75.0, 25.0)
            .apply_candles(&c, "close")?;

        assert_eq!(output.rows, 9);
        assert_eq!(output.cols, c.close.len());
        assert_eq!(output.combos.len(), 9);

        Ok(())
    }

    #[cfg(debug_assertions)]
    fn check_batch_no_poison(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let file = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let c = read_candles_from_csv(file)?;

        let output = PercentileNearestRankBatchBuilder::new()
            .kernel(kernel)
            .period_range(5, 20, 5)
            .percentage_range(10.0, 90.0, 20.0)
            .apply_candles(&c, "close")?;

        for &val in &output.values {
            if val.is_nan() {
                continue;
            }
            let bits = val.to_bits();
            if bits == 0x11111111_11111111
                || bits == 0x22222222_22222222
                || bits == 0x33333333_33333333
            {
                panic!(
                    "[{}] Found poison value {} (0x{:016X})",
                    test_name, val, bits
                );
            }
        }

        Ok(())
    }

    #[cfg(not(debug_assertions))]
    fn check_batch_no_poison(_test_name: &str, _kernel: Kernel) -> Result<(), Box<dyn Error>> {
        Ok(())
    }

    macro_rules! gen_batch_tests {
        ($fn_name:ident) => {
            paste::paste! {
                #[test] fn [<$fn_name _scalar>]() {
                    let _ = $fn_name(stringify!([<$fn_name _scalar>]), Kernel::ScalarBatch);
                }
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                #[test] fn [<$fn_name _avx2>]() {
                    let _ = $fn_name(stringify!([<$fn_name _avx2>]), Kernel::Avx2Batch);
                }
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                #[test] fn [<$fn_name _avx512>]() {
                    let _ = $fn_name(stringify!([<$fn_name _avx512>]), Kernel::Avx512Batch);
                }
                #[test] fn [<$fn_name _auto_detect>]() {
                    let _ = $fn_name(stringify!([<$fn_name _auto_detect>]), Kernel::Auto);
                }
            }
        };
    }

    gen_batch_tests!(check_batch_default_row);
    gen_batch_tests!(check_batch_sweep);
    gen_batch_tests!(check_batch_no_poison);

    #[test]
    fn test_percentile_nearest_rank_basic() {
        let data = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0];
        let params = PercentileNearestRankParams {
            length: Some(5),
            percentage: Some(50.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank(&input).unwrap();

        assert_eq!(result.values.len(), data.len());

        for i in 0..4 {
            assert!(result.values[i].is_nan());
        }

        assert_eq!(result.values[4], 3.0);
    }

    #[test]
    fn test_percentile_nearest_rank_empty_data() {
        let data = vec![];
        let params = PercentileNearestRankParams::default();
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank(&input);

        assert!(matches!(
            result,
            Err(PercentileNearestRankError::EmptyInputData)
        ));
    }

    #[test]
    fn test_percentile_nearest_rank_all_nan() {
        let data = vec![f64::NAN; 10];
        let params = PercentileNearestRankParams::default();
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank(&input);

        assert!(matches!(
            result,
            Err(PercentileNearestRankError::AllValuesNaN)
        ));
    }

    #[test]
    fn test_percentile_nearest_rank_invalid_percentage() {
        let data = vec![1.0; 20];
        let params = PercentileNearestRankParams {
            length: Some(5),
            percentage: Some(150.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank(&input);

        assert!(matches!(
            result,
            Err(PercentileNearestRankError::InvalidPercentage { .. })
        ));
    }

    #[test]
    fn test_percentile_nearest_rank_period_too_large() {
        let data = vec![1.0; 10];
        let params = PercentileNearestRankParams {
            length: Some(20),
            percentage: Some(50.0),
        };
        let input = PercentileNearestRankInput::from_slice(&data, params);
        let result = percentile_nearest_rank(&input);

        assert!(matches!(
            result,
            Err(PercentileNearestRankError::InvalidPeriod { .. })
        ));
    }

    #[test]
    fn test_percentile_nearest_rank_with_candles() {
        let close_data = vec![
            1.5, 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5, 9.5, 10.5, 11.5, 12.5, 13.5, 14.5, 15.5, 16.5,
            17.5, 18.5, 19.5, 20.5,
        ];
        let open_data = vec![1.0; 20];
        let high_data = vec![2.0; 20];
        let low_data = vec![0.5; 20];
        let volume_data = vec![100.0; 20];

        let mut hl2 = Vec::with_capacity(20);
        let mut hlc3 = Vec::with_capacity(20);
        let mut ohlc4 = Vec::with_capacity(20);
        let mut hlcc4 = Vec::with_capacity(20);

        for i in 0..20 {
            hl2.push((high_data[i] + low_data[i]) / 2.0);
            hlc3.push((high_data[i] + low_data[i] + close_data[i]) / 3.0);
            ohlc4.push((open_data[i] + high_data[i] + low_data[i] + close_data[i]) / 4.0);
            hlcc4.push((high_data[i] + low_data[i] + 2.0 * close_data[i]) / 4.0);
        }

        let candles = Candles {
            timestamp: vec![0; 20],
            open: open_data,
            high: high_data,
            low: low_data,
            close: close_data,
            volume: volume_data,
            fields: CandleFieldFlags {
                open: true,
                high: true,
                low: true,
                close: true,
                volume: true,
            },
            hl2,
            hlc3,
            ohlc4,
            hlcc4,
        };

        let params = PercentileNearestRankParams {
            length: Some(5),
            percentage: Some(50.0),
        };
        let input = PercentileNearestRankInput::from_candles(&candles, "close", params);
        let result = percentile_nearest_rank(&input).unwrap();

        assert_eq!(result.values.len(), 20);

        for i in 0..4 {
            assert!(result.values[i].is_nan());
        }

        assert_eq!(result.values[4], 3.5);
    }
}

#[cfg(feature = "python")]
pub fn register_percentile_nearest_rank_module(
    m: &Bound<'_, pyo3::types::PyModule>,
) -> PyResult<()> {
    m.add_function(wrap_pyfunction!(percentile_nearest_rank_py, m)?)?;
    m.add_function(wrap_pyfunction!(percentile_nearest_rank_batch_py, m)?)?;
    #[cfg(feature = "cuda")]
    {
        m.add_function(wrap_pyfunction!(
            percentile_nearest_rank_cuda_batch_dev_py,
            m
        )?)?;
        m.add_function(wrap_pyfunction!(
            percentile_nearest_rank_cuda_many_series_one_param_dev_py,
            m
        )?)?;
    }
    Ok(())
}

#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::cuda_available;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::percentile_nearest_rank_wrapper::CudaPercentileNearestRank;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::indicators::moving_averages::alma::DeviceArrayF32Py;

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "percentile_nearest_rank_cuda_batch_dev")]
#[pyo3(signature = (data_f32, length_range, percentage_range, device_id=0))]
pub fn percentile_nearest_rank_cuda_batch_dev_py<'py>(
    py: Python<'py>,
    data_f32: numpy::PyReadonlyArray1<'py, f32>,
    length_range: (usize, usize, usize),
    percentage_range: (f64, f64, f64),
    device_id: usize,
) -> PyResult<(DeviceArrayF32Py, Bound<'py, pyo3::types::PyDict>)> {
    use numpy::{IntoPyArray, PyArrayMethods};
    let slice_in = data_f32.as_slice()?;
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }
    let sweep = PercentileNearestRankBatchRange {
        length: length_range,
        percentage: percentage_range,
    };
    let (inner, ctx, dev_id, combos) = py.allow_threads(|| {
        let cuda = CudaPercentileNearestRank::new(device_id)
            .map_err(|e| PyValueError::new_err(e.to_string()))?;
        let ctx = cuda.context_arc();
        let dev_id = cuda.device_id();
        cuda.pnr_batch_dev(slice_in, &sweep)
            .map(|(inner, combos)| (inner, ctx, dev_id, combos))
            .map_err(|e| PyValueError::new_err(e.to_string()))
    })?;

    let dict = pyo3::types::PyDict::new(py);
    let lengths: Vec<u64> = combos
        .iter()
        .map(|c| c.length.unwrap_or(15) as u64)
        .collect();
    let percentages: Vec<f64> = combos
        .iter()
        .map(|c| c.percentage.unwrap_or(50.0))
        .collect();
    dict.set_item("lengths", lengths.into_pyarray(py))?;
    dict.set_item("percentages", percentages.into_pyarray(py))?;
    Ok((
        DeviceArrayF32Py {
            inner,
            _ctx: Some(ctx),
            device_id: Some(dev_id),
        },
        dict,
    ))
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "percentile_nearest_rank_cuda_many_series_one_param_dev")]
#[pyo3(signature = (data_tm_f32, cols, rows, length, percentage, device_id=0))]
pub fn percentile_nearest_rank_cuda_many_series_one_param_dev_py<'py>(
    py: Python<'py>,
    data_tm_f32: numpy::PyReadonlyArray1<'py, f32>,
    cols: usize,
    rows: usize,
    length: usize,
    percentage: f64,
    device_id: usize,
) -> PyResult<DeviceArrayF32Py> {
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }
    let slice_in = data_tm_f32.as_slice()?;
    let (inner, ctx, dev_id) = py.allow_threads(|| {
        let cuda = CudaPercentileNearestRank::new(device_id)
            .map_err(|e| PyValueError::new_err(e.to_string()))?;
        let ctx = cuda.context_arc();
        let dev_id = cuda.device_id();
        cuda.pnr_many_series_one_param_time_major_dev(slice_in, cols, rows, length, percentage)
            .map(|inner| (inner, ctx, dev_id))
            .map_err(|e| PyValueError::new_err(e.to_string()))
    })?;
    Ok(DeviceArrayF32Py {
        inner,
        _ctx: Some(ctx),
        device_id: Some(dev_id),
    })
}