nu_command/math/

median.rs

use crate::math::{
    avg::average,
    utils::{run_with_function_with_cell_paths, run_with_function_with_cell_paths_const},
};
use nu_engine::command_prelude::*;
use std::cmp::Ordering;

#[derive(Clone)]
pub struct MathMedian;

impl Command for MathMedian {
    fn name(&self) -> &str {
        "math median"
    }

    fn signature(&self) -> Signature {
        Signature::build("math median")
            .input_output_types(vec![
                (Type::List(Box::new(Type::Number)), Type::Number),
                (Type::List(Box::new(Type::Duration)), Type::Duration),
                (Type::List(Box::new(Type::Filesize)), Type::Filesize),
                (Type::Range, Type::Number),
                (Type::table(), Type::record()),
                (Type::record(), Type::record()),
            ])
            .allow_variants_without_examples(true)
            .rest(
                "columns",
                SyntaxShape::CellPath,
                "The cell-paths/columns to operate on.",
            )
            .category(Category::Math)
    }

    fn description(&self) -> &str {
        "Computes the median of a list of numbers."
    }

    fn search_terms(&self) -> Vec<&str> {
        vec!["middle", "statistics"]
    }

    fn is_const(&self) -> bool {
        true
    }

    fn run(
        &self,
        engine_state: &EngineState,
        stack: &mut Stack,
        call: &Call,
        input: PipelineData,
    ) -> Result<PipelineData, ShellError> {
        run_with_function_with_cell_paths(engine_state, stack, call, input, median)
    }

    fn run_const(
        &self,
        working_set: &StateWorkingSet,
        call: &Call,
        input: PipelineData,
    ) -> Result<PipelineData, ShellError> {
        run_with_function_with_cell_paths_const(working_set, call, input, median)
    }

    fn examples(&self) -> Vec<Example<'_>> {
        vec![
            Example {
                description: "Compute the median of a list of numbers.",
                example: "[3 8 9 12 12 15] | math median",
                result: Some(Value::test_float(10.5)),
            },
            Example {
                description: "Compute the medians of the columns of a table.",
                example: "[{a: 1 b: 3} {a: 2 b: -1} {a: -3 b: 5}] | math median",
                result: Some(Value::test_record(record! {
                    "a" => Value::test_int(1),
                    "b" => Value::test_int(3),
                })),
            },
            Example {
                description: "Find the median of a list of file sizes.",
                example: "[5KB 10MB 200B] | math median",
                result: Some(Value::test_filesize(5 * 1_000)),
            },
            Example {
                description: "Compute the median of list-valued columns in a record.",
                example: "{alice: [3 1 2], bob: [4 5 6]} | math median",
                result: Some(Value::test_record(record! {
                    "alice" => Value::test_int(2),
                    "bob" => Value::test_int(5),
                })),
            },
            Example {
                description: "Compute the median of a single column using a cell path.",
                example: "{alice: [3 1 2], bob: [4 5 6]} | math median alice",
                result: Some(Value::test_record(record! {
                    "alice" => Value::test_int(2),
                    "bob" => Value::list(
                        vec![Value::test_int(4), Value::test_int(5), Value::test_int(6)],
                        Span::test_data(),
                    ),
                })),
            },
        ]
    }
}

enum Pick {
    MedianAverage,
    Median,
}

pub fn median(values: &[Value], span: Span, head: Span) -> Result<Value, ShellError> {
    // Reject unsupported types up front so record columns error like other reducers.
    for value in values {
        match value {
            Value::Int { .. }
            | Value::Float { .. }
            | Value::Duration { .. }
            | Value::Filesize { .. } => {}
            Value::Error { error, .. } => return Err(*error.clone()),
            other => {
                return Err(ShellError::OnlySupportsThisInputType {
                    exp_input_type: crate::math::utils::NUMERIC_INPUT_TYPES.into(),
                    wrong_type: other.get_type().to_string(),
                    dst_span: head,
                    src_span: other.span(),
                });
            }
        }
    }

    let mut sorted = values
        .iter()
        .filter(|x| !x.as_float().is_ok_and(f64::is_nan))
        .collect::<Vec<_>>();

    if sorted.is_empty() {
        return Err(ShellError::UnsupportedInput {
            msg: "Empty input".to_string(),
            input: "value originates from here".into(),
            msg_span: head,
            input_span: span,
        });
    }

    sorted.sort_by(|a, b| a.partial_cmp(b).unwrap_or(Ordering::Equal));

    let take = if sorted.len().is_multiple_of(2) {
        Pick::MedianAverage
    } else {
        Pick::Median
    };

    match take {
        Pick::Median => {
            let idx = sorted.len() / 2;
            // Non-empty: middle element always exists.
            Ok(sorted[idx].to_owned().to_owned())
        }
        Pick::MedianAverage => {
            let idx_end = sorted.len() / 2;
            let idx_start = idx_end - 1;
            let left = sorted[idx_start].to_owned().to_owned();
            let right = sorted[idx_end].to_owned().to_owned();
            average(&[left, right], span, head)
        }
    }
}

#[cfg(test)]
mod test {
    use super::*;

    #[test]
    fn test_examples() -> nu_test_support::Result {
        nu_test_support::test().examples(MathMedian)
    }

    #[test]
    fn test_median_with_nan_values() {
        // Test case: [NaN, NaN, 1.0, 2.0, 3.0, 4.0]
        // values.len() = 6, sorted = [1.0, 2.0, 3.0, 4.0], sorted.len() = 4
        // Correct median of [1.0, 2.0, 3.0, 4.0] is (2.0 + 3.0) / 2 = 2.5
        // Bug: using values.len() would give idx_end = 3, returning avg(3.0, 4.0) = 3.5
        // Fix: using sorted.len() gives idx_end = 2, returning avg(2.0, 3.0) = 2.5
        let span = Span::test_data();
        let values = vec![
            Value::test_float(f64::NAN),
            Value::test_float(f64::NAN),
            Value::test_float(1.0),
            Value::test_float(2.0),
            Value::test_float(3.0),
            Value::test_float(4.0),
        ];

        let result = median(&values, span, span).unwrap();
        assert_eq!(result, Value::test_float(2.5));
    }
}