kestrel-chartkit 0.11.3

High-performance Rust technical analysis library for indicator math, market regime classification, composite scoring, and SVG visualization.
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
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198

//! Independent golden-reference and hand-calculation tests for out-of-sample data splits
//! (purging & embargo), frozen probability calibration, Brier Skill Score, and ECE.

mod common;

use kestrel_chartkit::evaluation::probability::{
    block_bootstrap_brier, compute_calibration_metrics, IsotonicCalibrator,
    ValidationExperimentManifest,
};
use kestrel_chartkit::evaluation::split::{split_trades_purged, PurgedSplitConfig, TradeSpan};

const GOLDEN: &str = include_str!("fixtures/golden_validation_probability.txt");

#[test]
fn test_golden_frozen_calibrator_invariant_to_test_data() {
    // Acceptance criterion 1: Testdaten ändern darf Training/Kalibrator nicht ändern.
    let train_data = vec![(20.0, false), (40.0, false), (60.0, true), (80.0, true)];
    let calibrator = IsotonicCalibrator::fit(&train_data).unwrap();

    let pred_at_50 = calibrator.predict(50.0);
    assert_eq!(
        pred_at_50, 0.5,
        "Midpoint between 40 (prob 0.0) and 60 (prob 1.0) must be 0.5"
    );

    // Feed arbitrary subsequent test datasets (e.g. all losses or all wins):
    let _test_set_a = [(50.0, false), (50.0, false), (50.0, false)];
    assert_eq!(
        calibrator.predict(50.0),
        pred_at_50,
        "Calibrator must remain strictly frozen regardless of test data"
    );

    let _test_set_b = [(50.0, true), (50.0, true), (50.0, true)];
    assert_eq!(
        calibrator.predict(50.0),
        pred_at_50,
        "Calibrator prediction cannot be influenced by test observations"
    );
}

#[test]
fn test_golden_purging_and_embargo_split() {
    // Acceptance criterion 2: kein Label greift über Split-Grenzen.
    let config = PurgedSplitConfig {
        train_start: 0,
        train_end: 100,
        test_start: 110, // 10 bars embargo
        test_end: 200,
        embargo_bars: 10,
    };

    let trades = vec![
        // Trade 1: safe inside train
        TradeSpan {
            id: 1,
            entry_bar: 10,
            exit_bar: 50,
        },
        // Trade 2: finishes inside embargo buffer (exit 105 < test_start 110)
        TradeSpan {
            id: 2,
            entry_bar: 85,
            exit_bar: 105,
        },
        // Trade 3: entered at 95, exits at 115 (extends past test_start 110) -> MUST BE PURGED!
        TradeSpan {
            id: 3,
            entry_bar: 95,
            exit_bar: 115,
        },
        // Trade 4: safe test trade
        TradeSpan {
            id: 4,
            entry_bar: 120,
            exit_bar: 160,
        },
    ];

    let split = split_trades_purged(&trades, &config).unwrap();
    assert_eq!(split.train_trade_ids, vec![1, 2]);
    assert_eq!(
        split.purged_trade_ids,
        vec![3],
        "Trade overlapping test set MUST be purged"
    );
    assert_eq!(split.test_trade_ids, vec![4]);
}

#[test]
fn test_golden_hand_calc_brier_score_and_brier_skill_score() {
    // Acceptance criterion 3: triviale konstante Prognose hat handgerechneten Brier Score.
    // 4 predictions: p = [0.8, 0.7, 0.4, 0.1]
    // 4 outcomes:    y = [true, true, false, false]
    let preds = vec![0.8, 0.7, 0.4, 0.1];
    let actuals = vec![true, true, false, false];

    // Hand calculation:
    // Squared errors:
    // (0.8 - 1.0)^2 = 0.04
    // (0.7 - 1.0)^2 = 0.09
    // (0.4 - 0.0)^2 = 0.16
    // (0.1 - 0.0)^2 = 0.01
    // Sum = 0.30 -> Brier Score = 0.30 / 4 = 0.0750
    //
    // Baseline constant prediction: 50% base rate (0.50)
    // Baseline squared errors: 4 * (0.5 - 1.0)^2 = 4 * 0.25 = 1.00 -> Baseline BS = 0.2500
    //
    // Brier Skill Score = 1.0 - (0.0750 / 0.2500) = 1.0 - 0.30 = 0.7000 (70% skill)
    //
    // Log-loss = -1/4 * [ln(0.8) + ln(0.7) + ln(0.6) + ln(0.9)], about 0.29900116; generated in
    // golden_validation_probability.txt.
    //
    // ECE across 2 bins ([0..0.5), [0.5..1.0]):
    // Bin 0: preds 0.4, 0.1 -> mean 0.25, actual winrate 0.0 -> gap 0.25 (weight 0.5)
    // Bin 1: preds 0.8, 0.7 -> mean 0.75, actual winrate 1.0 -> gap 0.25 (weight 0.5)
    // ECE = 0.5 * 0.25 + 0.5 * 0.25 = 0.25
    let metrics = compute_calibration_metrics(&preds, &actuals, 0.50, 2).unwrap();

    assert!((metrics.brier_score - 0.075).abs() < 1e-12);
    assert!((metrics.baseline_brier_score - 0.25).abs() < 1e-12);
    assert!((metrics.brier_skill_score - 0.70).abs() < 1e-12);
    common::assert_close(
        metrics.log_loss,
        common::golden_value(GOLDEN, "log_loss"),
        common::golden_value(GOLDEN, "validation_probability_tolerance"),
        "Log-Loss",
    );
    assert!((metrics.expected_calibration_error - 0.25).abs() < 1e-12);
    assert_eq!(metrics.sample_size, 4);
}

#[test]
fn test_golden_isotonic_calibrator_monotonicity() {
    // Non-monotone noisy inputs:
    // (10, 0), (20, 1), (30, 0), (40, 1), (50, 0), (60, 1)
    let noisy_data = vec![
        (10.0, false),
        (20.0, true),
        (30.0, false),
        (40.0, true),
        (50.0, false),
        (60.0, true),
    ];
    let calibrator = IsotonicCalibrator::fit(&noisy_data).unwrap();

    // PAVA must guarantee monotonicity everywhere
    let test_scores = [0.0, 15.0, 25.0, 35.0, 45.0, 55.0, 75.0];
    let mut prev_prob = 0.0;
    for &s in &test_scores {
        let prob = calibrator.predict(s);
        assert!(
            prob >= prev_prob - 1e-12,
            "Isotonic calibrator must be monotonically non-decreasing: at {s} got {prob} < {prev_prob}"
        );
        assert!((0.0..=1.0).contains(&prob));
        prev_prob = prob;
    }
}

#[test]
fn test_golden_low_sample_size_visibility() {
    // Acceptance criterion 4: geringe Stichprobe bleibt sichtbar.
    let small_preds = vec![0.6, 0.7, 0.8];
    let small_actuals = vec![false, true, true];
    let metrics = compute_calibration_metrics(&small_preds, &small_actuals, 0.5, 10).unwrap();

    assert_eq!(
        metrics.sample_size, 3,
        "Small sample size N=3 must be preserved explicitly"
    );

    let manifest = ValidationExperimentManifest {
        experiment_id: "exp_oos_001".to_string(),
        model_version: "v1.0.0".to_string(),
        train_range: (0, 100),
        test_range: (110, 150),
        embargo_bars: 10,
        train_sample_size: 20,
        test_sample_size: 3,
        metrics,
    };

    assert_eq!(manifest.test_sample_size, 3);
    assert_eq!(manifest.train_sample_size, 20);
}

#[test]
fn test_golden_block_bootstrap_brier() {
    let preds = vec![0.8, 0.7, 0.4, 0.1, 0.85, 0.65, 0.35, 0.15];
    let actuals = vec![true, true, false, false, true, true, false, false];

    let (mean_brier, p05, p95) = block_bootstrap_brier(&preds, &actuals, 2, 100, 42).unwrap();

    assert!(mean_brier > 0.0 && mean_brier < 0.25);
    assert!(p05 <= mean_brier);
    assert!(mean_brier <= p95);
}