infomeasure 0.1.0-alpha.1

Information theory measures and entropy calculations for Rust
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121

// SPDX-FileCopyrightText: 2025-2026 Carlson Büth <code@cbueth.de>
//
// SPDX-License-Identifier: MIT OR Apache-2.0

use ndarray::{Array1, Array2};
use statrs::function::gamma::digamma;

use crate::estimators::approaches::discrete::discrete_utils::reduce_joint_space_compact;
use crate::estimators::approaches::discrete::discrete_utils::{DiscreteDataset, rows_as_vec};
use crate::estimators::traits::{GlobalValue, JointEntropy, LocalValues, OptionalLocalValues};

/// ANSB (asymptotic NSB) entropy estimator for discrete data (natural log base).
///
/// Coincidence-based approximation to NSB that is appropriate in undersampled regimes.
/// Requires observed sample size N, observed K (or override), and uses coincidences Δ=N−K.
/// Returns NaN if inputs indicate inapplicability. Global-only.
///
/// Cross-entropy is not implemented for ANSB estimator.
/// The ANSB estimator is designed for single distribution entropy estimation
/// and cross-entropy creates a theoretical inconsistency.
///
/// Joint entropy is supported by reducing the joint space of multiple variables to a single
/// discrete representation before estimation.
///
/// Local values are not supported for the ANSB estimator.
/// The ANSB estimator is based on global statistics (coincidences) and
/// local values cannot be meaningfully extracted.
pub struct AnsbEntropy {
    dataset: DiscreteDataset,
    /// Optional override for support size K
    k_override: Option<usize>,
    /// Threshold for considering data sufficiently undersampled (ratio N/K)
    undersampled_threshold: f64,
}

impl AnsbEntropy {
    pub fn new(data: Array1<i32>, k_override: Option<usize>, undersampled_threshold: f64) -> Self {
        let dataset = DiscreteDataset::from_data(data);
        Self {
            dataset,
            k_override,
            undersampled_threshold,
        }
    }

    /// Build a vector of AnsbEntropy estimators, one per row of a 2D array.
    pub fn from_rows(
        data: Array2<i32>,
        k_override: Option<usize>,
        undersampled_threshold: f64,
    ) -> Vec<Self> {
        rows_as_vec(data)
            .into_iter()
            .map(|row| Self::new(row, k_override, undersampled_threshold))
            .collect()
    }
}

impl GlobalValue for AnsbEntropy {
    fn global_value(&self) -> f64 {
        let n = self.dataset.n;
        if n == 0 {
            return f64::NAN;
        }
        let k_obs = self.dataset.k;
        let k = self.k_override.unwrap_or(k_obs);
        if k == 0 {
            return f64::NAN;
        }

        // Check if data is sufficiently undersampled
        let ratio = n as f64 / k as f64;
        if ratio > self.undersampled_threshold {
            println!(
                "Warning: Data is not sufficiently undersampled (N/K = {:.3} > {:.3}), so calculation may diverge...",
                ratio, self.undersampled_threshold
            );
        }

        let coincidences = (n as i64) - (k as i64);
        if coincidences <= 0 {
            return f64::NAN;
        }

        // (γ - ln 2) + 2 ln N - ψ(Δ)
        const EULER_GAMMA: f64 = 0.577_215_664_901_532_9;

        (EULER_GAMMA - 2.0_f64.ln()) + 2.0 * (n as f64).ln() - digamma(coincidences as f64)
    }
}

impl LocalValues for AnsbEntropy {
    fn local_values(&self) -> Array1<f64> {
        Array1::zeros(0)
    }
}

impl JointEntropy for AnsbEntropy {
    type Source = Array1<i32>;
    type Params = (Option<usize>, f64); // k_override, undersampled_threshold

    fn joint_entropy(series: &[Self::Source], params: Self::Params) -> f64 {
        if series.is_empty() {
            return 0.0;
        }
        let joint_codes = reduce_joint_space_compact(series);
        let disc = AnsbEntropy::new(joint_codes, params.0, params.1);
        disc.global_value()
    }
}

impl OptionalLocalValues for AnsbEntropy {
    fn supports_local(&self) -> bool {
        false
    }
    fn local_values_opt(&self) -> Result<Array1<f64>, &'static str> {
        Err(
            "Local values are not supported for ANSB estimator as it averages over Dirichlet priors.",
        )
    }
}