reqsketch 0.1.3

Relative Error Quantiles sketch
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

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

    #[test]
    fn test_theoretical_error_bounds() -> Result<()> {
        let mut sketch = ReqSketch::new();
        let n = 50_000;

        // Add uniformly distributed values
        for i in 0..n {
            sketch.update(i as f64);
        }

        println!("=== Rust REQ Error Bounds Validation ===");
        println!("Using theoretical error bounds from C++ reference implementation");
        println!("");

        // Comprehensive quantile coverage - from very low to very high
        let test_ranks = [
            0.01, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.4, 0.5, 0.6, 0.7, 0.75,
            0.8, 0.85, 0.9, 0.92, 0.95, 0.97, 0.98, 0.99, 0.995, 0.999
        ];

        println!("Rank\tQuantile\tTrue Val\tRank Error\t3σ Bound\tWithin Bound");
        println!("----\t--------\t--------\t----------\t--------\t------------");

        for &rank in &test_ranks {
            // Test 1: For a known quantile, check if our rank estimate is within bounds
            let true_quantile = rank * (n - 1) as f64;
            let estimated_rank = sketch.rank(&true_quantile, SearchCriteria::Inclusive)?;

            // Get theoretical error bounds at 3 standard deviations (99.7% confidence)
            let lower_bound = sketch.get_rank_lower_bound(rank, 3);
            let upper_bound = sketch.get_rank_upper_bound(rank, 3);
            let theoretical_error = ((rank - lower_bound).max(upper_bound - rank)).max(0.0);

            let rank_error = (estimated_rank - rank).abs();
            let within_bounds = estimated_rank >= lower_bound && estimated_rank <= upper_bound;

            println!("{}\t\t{:.0}\t\t{:.1}\t\t{:.4}\t\t±{:.2}%\t\t{}",
                     rank, true_quantile, true_quantile, rank_error,
                     theoretical_error * 100.0,
                     if within_bounds { "YES" } else { "NO" });

            // Log violations to see the full pattern first
            if !within_bounds {
                let error_pct = (estimated_rank - rank).abs() / rank * 100.0;
                eprintln!("VIOLATION: Rank {} estimate {:.6} outside bounds [{:.6}, {:.6}], error {:.2}%",
                         rank, estimated_rank, lower_bound, upper_bound, error_pct);
            }

            // Assert bounds compliance now that bounds calculation is fixed
            assert!(within_bounds,
                "Estimated rank {:.6} for true quantile {:.1} (rank {}) is outside 3-sigma bounds [{:.6}, {:.6}]. Error: {:.2}%. This violates the fundamental REQ sketch guarantee.",
                estimated_rank, true_quantile, rank, lower_bound, upper_bound, (estimated_rank - rank).abs() / rank * 100.0);
        }

        println!("\n✓ All quantile estimates are within 3-sigma bounds as required");
        Ok(())
    }

    #[test]
    fn test_error_bounds_across_k_values() -> Result<()> {
        println!("=== Error Bounds vs K Parameter ===");
        println!("Comparing theoretical accuracy for different k values");
        println!("");

        let k_values = [8, 12, 16, 24, 32];
        let test_rank = 0.25;

        println!("K\t1σ Error\t2σ Error\t3σ Error");
        println!("-\t--------\t--------\t--------");

        for &k in &k_values {
            let mut sketch = ReqSketch::builder().k(k)?.build()?;

            // Add enough data to trigger estimation mode
            for i in 0..10000 {
                sketch.update(i as f64);
            }

            let error_1sigma = (test_rank - sketch.get_rank_lower_bound(test_rank, 1)).max(
                sketch.get_rank_upper_bound(test_rank, 1) - test_rank);
            let error_2sigma = (test_rank - sketch.get_rank_lower_bound(test_rank, 2)).max(
                sketch.get_rank_upper_bound(test_rank, 2) - test_rank);
            let error_3sigma = (test_rank - sketch.get_rank_lower_bound(test_rank, 3)).max(
                sketch.get_rank_upper_bound(test_rank, 3) - test_rank);

            println!("{}\t{:.2}%\t\t{:.2}%\t\t{:.2}%",
                     k, error_1sigma * 100.0, error_2sigma * 100.0, error_3sigma * 100.0);
        }

        println!("\n✓ Error bounds decrease as k increases, as expected");
        Ok(())
    }

    #[test]
    fn test_hra_vs_lra_error_bounds() -> Result<()> {
        println!("=== HRA vs LRA Error Bounds Comparison ===");
        println!("High Rank Accuracy should have better bounds for high ranks");
        println!("Low Rank Accuracy should have better bounds for low ranks");
        println!("");

        let test_ranks = [0.05, 0.25, 0.5, 0.75, 0.95];

        println!("Rank\tHRA Error\tLRA Error\tBetter Mode");
        println!("----\t---------\t---------\t-----------");

        for &rank in &test_ranks {
            let mut hra_sketch = ReqSketch::builder()
                .rank_accuracy(RankAccuracy::HighRank)
                .build()?;
            let mut lra_sketch = ReqSketch::builder()
                .rank_accuracy(RankAccuracy::LowRank)
                .build()?;

            // Add data to both sketches
            for i in 0..10000 {
                let val = i as f64;
                hra_sketch.update(val);
                lra_sketch.update(val);
            }

            let hra_error = (rank - hra_sketch.get_rank_lower_bound(rank, 2)).max(
                hra_sketch.get_rank_upper_bound(rank, 2) - rank);
            let lra_error = (rank - lra_sketch.get_rank_lower_bound(rank, 2)).max(
                lra_sketch.get_rank_upper_bound(rank, 2) - rank);

            let better = if hra_error < lra_error { "HRA" } else { "LRA" };

            println!("{}\t{:.2}%\t\t{:.2}%\t\t{}",
                     rank, hra_error * 100.0, lra_error * 100.0, better);

            // Verify the expected behavior
            if rank >= 0.75 {
                assert!(hra_error <= lra_error,
                       "HRA should have better bounds for high ranks, but got HRA: {:.4}%, LRA: {:.4}%",
                       hra_error * 100.0, lra_error * 100.0);
            } else if rank <= 0.25 {
                assert!(lra_error <= hra_error,
                       "LRA should have better bounds for low ranks, but got HRA: {:.4}%, LRA: {:.4}%",
                       hra_error * 100.0, lra_error * 100.0);
            }
        }

        println!("\n✓ HRA and LRA modes show expected accuracy patterns");
        Ok(())
    }

    #[test]
    fn test_exact_mode_error_bounds() -> Result<()> {
        println!("=== Exact Mode Error Bounds ===");
        println!("Small sketches should have exact bounds (0% error)");
        println!("");

        let mut sketch = ReqSketch::new();

        // Add only a few values to stay in exact mode
        for i in 0..10 {
            sketch.update(i as f64);
        }

        assert!(!sketch.is_estimation_mode(), "Should be in exact mode");

        let test_ranks = [0.1, 0.25, 0.5, 0.75, 0.9];

        for &rank in &test_ranks {
            let lower = sketch.get_rank_lower_bound(rank, 2);
            let upper = sketch.get_rank_upper_bound(rank, 2);

            println!("Rank {}: bounds [{:.6}, {:.6}]", rank, lower, upper);

            // In exact mode, bounds should be exact (or very close due to discretization)
            let error = (upper - lower).max(0.0);
            assert!(error < 0.01, "Exact mode should have minimal error bounds, got {:.4}", error);
        }

        println!("\n✓ Exact mode provides tight error bounds as expected");
        Ok(())
    }
}