Files
IronCalc/base/src/functions/statistical/fisher.rs
2025-11-26 00:30:37 +01:00

419 lines
13 KiB
Rust

use statrs::distribution::{Continuous, ContinuousCDF, FisherSnedecor};
use crate::expressions::types::CellReferenceIndex;
use crate::functions::statistical::t_dist::sample_var;
use crate::{
calc_result::CalcResult, expressions::parser::Node, expressions::token::Error, model::Model,
};
impl Model {
// FISHER(x) = 0.5 * ln((1 + x) / (1 - x))
pub(crate) fn fn_fisher(&mut self, args: &[Node], cell: CellReferenceIndex) -> CalcResult {
if args.len() != 1 {
return CalcResult::new_args_number_error(cell);
}
let x = match self.get_number_no_bools(&args[0], cell) {
Ok(f) => f,
Err(e) => return e,
};
if x <= -1.0 || x >= 1.0 {
return CalcResult::Error {
error: Error::NUM,
origin: cell,
message: "x must be between -1 and 1 (exclusive) in FISHER".to_string(),
};
}
let ratio = (1.0 + x) / (1.0 - x);
let result = 0.5 * ratio.ln();
if result.is_nan() || result.is_infinite() {
return CalcResult::Error {
error: Error::NUM,
origin: cell,
message: "Invalid result for FISHER".to_string(),
};
}
CalcResult::Number(result)
}
// FISHERINV(y) = (e^(2y) - 1) / (e^(2y) + 1) = tanh(y)
pub(crate) fn fn_fisher_inv(&mut self, args: &[Node], cell: CellReferenceIndex) -> CalcResult {
if args.len() != 1 {
return CalcResult::new_args_number_error(cell);
}
let y = match self.get_number_no_bools(&args[0], cell) {
Ok(f) => f,
Err(e) => return e,
};
// Use tanh directly to avoid overflow from exp(2y)
let result = y.tanh();
if result.is_nan() || result.is_infinite() {
return CalcResult::Error {
error: Error::NUM,
origin: cell,
message: "Invalid result for FISHERINV".to_string(),
};
}
CalcResult::Number(result)
}
// F.DIST(x, deg_freedom1, deg_freedom2, cumulative)
pub(crate) fn fn_f_dist(&mut self, args: &[Node], cell: CellReferenceIndex) -> CalcResult {
if args.len() != 4 {
return CalcResult::new_args_number_error(cell);
}
let x = match self.get_number_no_bools(&args[0], cell) {
Ok(f) => f,
Err(e) => return e,
};
let df1 = match self.get_number_no_bools(&args[1], cell) {
Ok(f) => f.trunc(),
Err(e) => return e,
};
let df2 = match self.get_number_no_bools(&args[2], cell) {
Ok(f) => f.trunc(),
Err(e) => return e,
};
let cumulative = match self.get_boolean(&args[3], cell) {
Ok(b) => b,
Err(e) => return e,
};
// Excel domain checks
if x < 0.0 {
return CalcResult::new_error(Error::NUM, cell, "x must be >= 0 in F.DIST".to_string());
}
if df1 < 1.0 || df2 < 1.0 {
return CalcResult::new_error(
Error::NUM,
cell,
"degrees of freedom must be >= 1 in F.DIST".to_string(),
);
}
let dist = match FisherSnedecor::new(df1, df2) {
Ok(d) => d,
Err(_) => {
return CalcResult::new_error(
Error::NUM,
cell,
"Invalid parameters for F distribution".to_string(),
)
}
};
let result = if cumulative { dist.cdf(x) } else { dist.pdf(x) };
if result.is_nan() || result.is_infinite() {
return CalcResult::new_error(
Error::NUM,
cell,
"Invalid result for F.DIST".to_string(),
);
}
CalcResult::Number(result)
}
pub(crate) fn fn_f_dist_rt(&mut self, args: &[Node], cell: CellReferenceIndex) -> CalcResult {
// F.DIST.RT(x, deg_freedom1, deg_freedom2)
if args.len() != 3 {
return CalcResult::new_args_number_error(cell);
}
let x = match self.get_number_no_bools(&args[0], cell) {
Ok(f) => f,
Err(e) => return e,
};
let df1 = match self.get_number_no_bools(&args[1], cell) {
Ok(f) => f.trunc(),
Err(e) => return e,
};
let df2 = match self.get_number_no_bools(&args[2], cell) {
Ok(f) => f.trunc(),
Err(e) => return e,
};
if x < 0.0 {
return CalcResult::new_error(
Error::NUM,
cell,
"x must be >= 0 in F.DIST.RT".to_string(),
);
}
if df1 < 1.0 || df2 < 1.0 {
return CalcResult::new_error(
Error::NUM,
cell,
"degrees of freedom must be >= 1 in F.DIST.RT".to_string(),
);
}
let dist = match FisherSnedecor::new(df1, df2) {
Ok(d) => d,
Err(_) => {
return CalcResult::new_error(
Error::NUM,
cell,
"Invalid parameters for F distribution".to_string(),
)
}
};
// Right-tail probability: P(F > x) = 1 - CDF(x)
let result = 1.0 - dist.cdf(x);
if result.is_nan() || result.is_infinite() || result < 0.0 {
return CalcResult::new_error(
Error::NUM,
cell,
"Invalid result for F.DIST.RT".to_string(),
);
}
CalcResult::Number(result)
}
// F.INV(probability, deg_freedom1, deg_freedom2)
pub(crate) fn fn_f_inv(&mut self, args: &[Node], cell: CellReferenceIndex) -> CalcResult {
if args.len() != 3 {
return CalcResult::new_args_number_error(cell);
}
let p = match self.get_number_no_bools(&args[0], cell) {
Ok(f) => f,
Err(e) => return e,
};
let df1 = match self.get_number_no_bools(&args[1], cell) {
Ok(f) => f.trunc(),
Err(e) => return e,
};
let df2 = match self.get_number_no_bools(&args[2], cell) {
Ok(f) => f.trunc(),
Err(e) => return e,
};
// probability < 0 or > 1 → #NUM!
if !(0.0..=1.0).contains(&p) {
return CalcResult::new_error(
Error::NUM,
cell,
"probability must be in [0,1] in F.INV".to_string(),
);
}
if df1 < 1.0 || df2 < 1.0 {
return CalcResult::new_error(
Error::NUM,
cell,
"degrees of freedom must be >= 1 in F.INV".to_string(),
);
}
let dist = match FisherSnedecor::new(df1, df2) {
Ok(d) => d,
Err(_) => {
return CalcResult::new_error(
Error::NUM,
cell,
"Invalid parameters for F distribution".to_string(),
)
}
};
let x = dist.inverse_cdf(p);
if x.is_nan() || x.is_infinite() || x < 0.0 {
return CalcResult::new_error(Error::NUM, cell, "Invalid result for F.INV".to_string());
}
CalcResult::Number(x)
}
// F.INV.RT(probability, deg_freedom1, deg_freedom2)
pub(crate) fn fn_f_inv_rt(&mut self, args: &[Node], cell: CellReferenceIndex) -> CalcResult {
if args.len() != 3 {
return CalcResult::new_args_number_error(cell);
}
let p = match self.get_number_no_bools(&args[0], cell) {
Ok(f) => f,
Err(e) => return e,
};
let df1 = match self.get_number_no_bools(&args[1], cell) {
Ok(f) => f.trunc(),
Err(e) => return e,
};
let df2 = match self.get_number_no_bools(&args[2], cell) {
Ok(f) => f.trunc(),
Err(e) => return e,
};
if p <= 0.0 || p > 1.0 {
return CalcResult::new_error(
Error::NUM,
cell,
"probability must be in (0,1] in F.INV.RT".to_string(),
);
}
if df1 < 1.0 || df2 < 1.0 {
return CalcResult::new_error(
Error::NUM,
cell,
"degrees of freedom must be >= 1 in F.INV.RT".to_string(),
);
}
let dist = match FisherSnedecor::new(df1, df2) {
Ok(d) => d,
Err(_) => {
return CalcResult::new_error(
Error::NUM,
cell,
"Invalid parameters for F distribution".to_string(),
)
}
};
// p is right-tail: p = P(F > x) = 1 - CDF(x)
let x = dist.inverse_cdf(1.0 - p);
if x.is_nan() || x.is_infinite() || x < 0.0 {
return CalcResult::new_error(
Error::NUM,
cell,
"Invalid result for F.INV.RT".to_string(),
);
}
CalcResult::Number(x)
}
// F.TEST(array1, array2)
pub(crate) fn fn_f_test(&mut self, args: &[Node], cell: CellReferenceIndex) -> CalcResult {
if args.len() != 2 {
return CalcResult::new_args_number_error(cell);
}
let values1_opts = match self.evaluate_node_in_context(&args[0], cell) {
CalcResult::Range { left, right } => match self.values_from_range(left, right) {
Ok(v) => v,
Err(error) => return error,
},
CalcResult::Array(a) => match self.values_from_array(a) {
Ok(v) => v,
Err(error) => {
return CalcResult::new_error(
Error::VALUE,
cell,
format!("Error in first array: {:?}", error),
);
}
},
_ => {
return CalcResult::new_error(
Error::VALUE,
cell,
"First argument must be a range or array".to_string(),
);
}
};
// Get second sample as Vec<Option<f64>>
let values2_opts = match self.evaluate_node_in_context(&args[1], cell) {
CalcResult::Range { left, right } => match self.values_from_range(left, right) {
Ok(v) => v,
Err(error) => return error,
},
CalcResult::Array(a) => match self.values_from_array(a) {
Ok(v) => v,
Err(error) => {
return CalcResult::new_error(
Error::VALUE,
cell,
format!("Error in second array: {:?}", error),
);
}
},
_ => {
return CalcResult::new_error(
Error::VALUE,
cell,
"Second argument must be a range or array".to_string(),
);
}
};
let values1: Vec<f64> = values1_opts.into_iter().flatten().collect();
let values2: Vec<f64> = values2_opts.into_iter().flatten().collect();
let n1 = values1.len();
let n2 = values2.len();
// If fewer than 2 numeric values in either sample -> #DIV/0!
if n1 < 2 || n2 < 2 {
return CalcResult::new_error(
Error::DIV,
cell,
"F.TEST requires at least two numeric values in each sample".to_string(),
);
}
let v1 = sample_var(&values1);
let v2 = sample_var(&values2);
if v1 <= 0.0 || v2 <= 0.0 {
return CalcResult::new_error(
Error::DIV,
cell,
"Variance of one sample is zero in F.TEST".to_string(),
);
}
// F ratio: larger variance / smaller variance
let mut f = v1 / v2;
let mut df1 = (n1 - 1) as f64;
let mut df2 = (n2 - 1) as f64;
if f < 1.0 {
f = 1.0 / f;
std::mem::swap(&mut df1, &mut df2);
}
let dist = match FisherSnedecor::new(df1, df2) {
Ok(d) => d,
Err(_) => {
return CalcResult::new_error(
Error::NUM,
cell,
"Invalid parameters for F distribution in F.TEST".to_string(),
);
}
};
// One-tailed right-tail probability
let tail = 1.0 - dist.cdf(f);
// F.TEST is two-tailed: p = 2 * tail (with F >= 1)
let mut p = 2.0 * tail;
// Clamp tiny FP noise
if p < 0.0 && p > -1e-15 {
p = 0.0;
}
if p > 1.0 && p < 1.0 + 1e-15 {
p = 1.0;
}
CalcResult::Number(p)
}
}