uqa-planner 0.4.9

Cost model, cardinality, DPccp join enumeration, optimizer rewrites
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

//
// Unified Query Algebra
//
// Copyright (c) 2023-2026 Cognica, Inc.
//

//! Core `OperatorTree` cardinality dispatch and intersection damping.

use super::{
    column_entropy, entropy_cardinality_lower_bound, mutual_information_estimate,
    CardinalityEstimator, GraphStoreSampler, IndexStats, OperatorTree,
};

impl CardinalityEstimator {
    /// Estimate the cardinality of `op` against an inverted index described
    /// by `stats`.
    pub fn estimate(&self, op: &OperatorTree, stats: &IndexStats) -> f64 {
        let n = if stats.total_docs > 0 {
            stats.total_docs as f64
        } else {
            1.0
        };

        match op {
            OperatorTree::Empty => 0.0,
            OperatorTree::Term { query, field, .. } | OperatorTree::Phrase { query, field, .. } => {
                let field_name = field.as_deref().unwrap_or("_default");
                stats.doc_freq(field_name, query) as f64
            }
            OperatorTree::VectorSimilarity { threshold, .. } => {
                n * Self::vector_selectivity(f64::from(*threshold), stats.dimensions)
            }
            OperatorTree::KNN { k, .. } => (*k as f64).min(n),
            OperatorTree::Filter {
                field,
                predicate,
                source,
            } => {
                let input = source
                    .as_deref()
                    .map_or(n, |source| self.estimate(source, stats));
                input * self.filter_selectivity(field, predicate, n)
            }
            OperatorTree::Score { source, .. } => self.estimate(source, stats),
            OperatorTree::BayesianScore { source, .. } => self.estimate(source, stats),
            OperatorTree::Intersect(ops) => self.estimate_intersect(ops, stats, n),
            OperatorTree::Union(ops) => {
                let child_cards: f64 = ops.iter().map(|o| self.estimate(o, stats)).sum();
                n.min(child_cards)
            }
            OperatorTree::Complement(inner) => {
                let inner_card = self.estimate(inner, stats);
                (n - inner_card).max(0.0)
            }
            _ => self.estimate_cross_paradigm(op, stats, n),
        }
    }

    /// Backward-compat alias used by [`crate::query_optimizer`]. Builds
    /// an [`IndexStats`] with `total_docs = row_count.unwrap_or(1000)`
    /// and routes through [`Self::estimate`].
    pub fn estimate_operator(&self, op: &OperatorTree, row_count: Option<u64>) -> f64 {
        let stats = IndexStats::new(row_count.unwrap_or(1_000));
        self.estimate(op, &stats)
    }

    fn estimate_intersect(&self, ops: &[OperatorTree], stats: &IndexStats, n: f64) -> f64 {
        let mut child_cards: Vec<f64> = ops.iter().map(|o| self.estimate(o, stats)).collect();
        if child_cards.is_empty() {
            return 0.0;
        }
        child_cards.sort_by(f64::total_cmp);

        let damping = self.intersection_damping(ops);
        let mut result = child_cards[0];
        for card in &child_cards[1..] {
            let sel = if n > 0.0 { card / n } else { 1.0 };
            result *= sel.powf(damping);
        }

        // Apply entropy-based lower bound (Paper 1, Section 7) when
        // column stats are available.
        if !self.column_stats.is_empty() {
            let mut entropies: Vec<f64> = Vec::new();
            for op_item in ops {
                if let OperatorTree::Filter { field, .. } = op_item {
                    if let Some(cs) = self.column_stats.get(field) {
                        entropies.push(column_entropy(cs));
                    }
                }
            }
            if !entropies.is_empty() {
                let lb = entropy_cardinality_lower_bound(n, &entropies);
                result = result.max(lb);
            }
        }

        result.max(1.0)
    }

    /// Choose a damping exponent based on predicate correlation.
    fn intersection_damping(&self, ops: &[OperatorTree]) -> f64 {
        let fields: Vec<&str> = ops
            .iter()
            .filter_map(|o| match o {
                OperatorTree::Filter { field, .. } => Some(field.as_str()),
                _ => None,
            })
            .collect();

        if fields.len() < 2 {
            return 0.5;
        }

        let unique: std::collections::BTreeSet<&str> = fields.iter().copied().collect();
        if unique.len() == 1 {
            return 0.1;
        }

        if !self.column_stats.is_empty() && fields.len() >= 2 {
            let cs_a = self.column_stats.get(fields[0]);
            let cs_b = self.column_stats.get(fields[1]);
            if let (Some(a), Some(b)) = (cs_a, cs_b) {
                let mi = mutual_information_estimate(a, b, 0.1);
                if mi > 1.0 {
                    return 0.2;
                }
                if mi > 0.5 {
                    return 0.3;
                }
            }
        }

        0.5
    }
}