arcweight 0.3.0

A high-performance, modular library for weighted finite state transducers with comprehensive examples and benchmarks
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

//! Prune operation benchmarks
//!
//! These benchmarks measure the performance of FST pruning operations
//! across different FST sizes, weight thresholds, and structures.

use arcweight::algorithms::PruneConfig;
use arcweight::prelude::*;
use criterion::{criterion_group, criterion_main, Criterion};
use std::hint::black_box;

/// Create a linear FST with the specified number of states
fn create_linear_fst(size: usize) -> VectorFst<TropicalWeight> {
    let mut fst = VectorFst::new();

    if size == 0 {
        return fst;
    }

    // Add states
    let states: Vec<_> = (0..size).map(|_| fst.add_state()).collect();

    // Set start and final states
    fst.set_start(states[0]);
    fst.set_final(states[size - 1], TropicalWeight::one());

    // Add transitions in a chain
    for i in 0..size - 1 {
        fst.add_arc(
            states[i],
            Arc::new(
                (i % 26 + 97) as u32,
                (i % 26 + 97) as u32,
                TropicalWeight::new(1.0),
                states[i + 1],
            ),
        );
    }

    fst
}

/// Create a branching FST where each state has multiple outgoing arcs with varying weights
fn create_branching_fst(states: usize, branches: usize) -> VectorFst<TropicalWeight> {
    let mut fst = VectorFst::new();

    if states == 0 {
        return fst;
    }

    // Add states
    let state_ids: Vec<_> = (0..states).map(|_| fst.add_state()).collect();

    // Set start state
    fst.set_start(state_ids[0]);
    fst.set_final(state_ids[states - 1], TropicalWeight::one());

    // Add branching transitions with varying weights
    for i in 0..states - 1 {
        for j in 0..branches.min(states - i - 1) {
            let target_state = state_ids[i + j + 1];
            fst.add_arc(
                state_ids[i],
                Arc::new(
                    ((i + j) % 26 + 97) as u32,
                    ((i + j) % 26 + 97) as u32,
                    TropicalWeight::new((j + 1) as f32 * 2.0), // Varying weights
                    target_state,
                ),
            );
        }
    }

    fst
}

fn bench_prune_linear(c: &mut Criterion) {
    let mut group = c.benchmark_group("prune_linear");

    for size in [100, 1000, 5000].iter() {
        let fst = create_linear_fst(*size);

        group.bench_function(format!("default_config_{size}"), |b| {
            b.iter(|| {
                let result: VectorFst<TropicalWeight> =
                    prune(black_box(&fst), PruneConfig::default()).unwrap();
                black_box(result)
            })
        });

        let config = PruneConfig {
            weight_threshold: 10.0,
            state_threshold: None,
            npath: None,
            use_forward_backward: false,
            delta: 1e-6,
        };

        group.bench_function(format!("threshold_10_{size}"), |b| {
            b.iter(|| {
                let result: VectorFst<TropicalWeight> =
                    prune(black_box(&fst), config.clone()).unwrap();
                black_box(result)
            })
        });
    }

    group.finish();
}

fn bench_prune_branching(c: &mut Criterion) {
    let mut group = c.benchmark_group("prune_branching");

    for size in [50, 200, 500, 1000].iter() {
        let fst = create_branching_fst(*size, 3);

        group.bench_function(format!("default_config_{size}"), |b| {
            b.iter(|| {
                let result: VectorFst<TropicalWeight> =
                    prune(black_box(&fst), PruneConfig::default()).unwrap();
                black_box(result)
            })
        });

        let config = PruneConfig {
            weight_threshold: 5.0,
            state_threshold: Some(size / 2),
            npath: None,
            use_forward_backward: false,
            delta: 1e-6,
        };

        group.bench_function(format!("threshold_5_states_{size}"), |b| {
            b.iter(|| {
                let result: VectorFst<TropicalWeight> =
                    prune(black_box(&fst), config.clone()).unwrap();
                black_box(result)
            })
        });
    }

    group.finish();
}

criterion_group!(
    benches,
    bench_prune_linear,
    bench_prune_branching
);
criterion_main!(benches);