rill-ml 0.7.2

Lightweight, serializable online machine learning for Rust applications and streaming data.
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
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318

//! Integration tests: bandit learning behavior.
//!
//! These tests verify that each bandit algorithm can learn to identify the
//! best arm in a controlled setting, and that error handling and reset
//! behavior work correctly.

use rand::SeedableRng;
use rand_chacha::ChaCha8Rng;

use rill_ml::bandit::{
    Bandit, ContextualBandit, EpsilonGreedy, EpsilonGreedyConfig, LinUcb, LinUcbConfig,
    ThompsonConfig, ThompsonSampling, Ucb1, Ucb1Config,
};
use rill_ml::error::RillError;

/// Simulate a reward for a given arm. Arm 0 has the highest mean reward.
fn reward_for_arm(arm: usize, rng: &mut ChaCha8Rng) -> f64 {
    match arm {
        0 => 0.8 + rand::Rng::gen_range(rng, -0.05..0.05),
        1 => 0.3 + rand::Rng::gen_range(rng, -0.05..0.05),
        _ => 0.5 + rand::Rng::gen_range(rng, -0.05..0.05),
    }
}

/// Clamp a value into [0, 1] for ThompsonSampling (which requires [0, 1] rewards).
fn clamp_01(v: f64) -> f64 {
    v.clamp(0.0, 1.0)
}

#[test]
fn epsilon_greedy_finds_best_arm() {
    let mut rng = ChaCha8Rng::seed_from_u64(42);
    let mut bandit = EpsilonGreedy::new(
        3,
        EpsilonGreedyConfig {
            epsilon: 0.1,
            decay: 1.0,
            min_epsilon: 0.1,
        },
    )
    .unwrap();

    for _ in 0..1000 {
        let arm = bandit.select(&mut rng).unwrap();
        let reward = reward_for_arm(arm, &mut rng);
        bandit.update(arm, reward).unwrap();
    }

    let s0 = bandit.arm_stats(0).unwrap();
    let s1 = bandit.arm_stats(1).unwrap();
    let s2 = bandit.arm_stats(2).unwrap();
    assert!(
        s0.pulls > s1.pulls,
        "arm 0 should be pulled more than arm 1"
    );
    assert!(
        s0.pulls > s2.pulls,
        "arm 0 should be pulled more than arm 2"
    );
    assert!(
        s0.mean_reward > s1.mean_reward,
        "arm 0 should have higher mean reward"
    );
}

#[test]
fn ucb1_finds_best_arm() {
    let mut rng = ChaCha8Rng::seed_from_u64(42);
    let mut bandit = Ucb1::new(3, Ucb1Config::default()).unwrap();

    for _ in 0..1000 {
        let arm = bandit.select(&mut rng).unwrap();
        let reward = reward_for_arm(arm, &mut rng);
        bandit.update(arm, reward).unwrap();
    }

    let s0 = bandit.arm_stats(0).unwrap();
    let s1 = bandit.arm_stats(1).unwrap();
    let s2 = bandit.arm_stats(2).unwrap();
    assert!(
        s0.pulls > s1.pulls,
        "arm 0 should be pulled more than arm 1"
    );
    assert!(
        s0.pulls > s2.pulls,
        "arm 0 should be pulled more than arm 2"
    );
}

#[test]
fn thompson_sampling_finds_best_arm() {
    let mut rng = ChaCha8Rng::seed_from_u64(42);
    let mut bandit = ThompsonSampling::new(3, ThompsonConfig::default()).unwrap();

    for _ in 0..1000 {
        let arm = bandit.select(&mut rng).unwrap();
        let reward = clamp_01(reward_for_arm(arm, &mut rng));
        bandit.update(arm, reward).unwrap();
    }

    let s0 = bandit.arm_stats(0).unwrap();
    let s1 = bandit.arm_stats(1).unwrap();
    let s2 = bandit.arm_stats(2).unwrap();
    assert!(
        s0.pulls > s1.pulls,
        "arm 0 should be pulled more than arm 1"
    );
    assert!(
        s0.pulls > s2.pulls,
        "arm 0 should be pulled more than arm 2"
    );
}

#[test]
fn linucb_learns_contextual_policy() {
    let mut rng = ChaCha8Rng::seed_from_u64(42);
    let mut bandit = LinUcb::new(LinUcbConfig {
        alpha: 1.0,
        arm_count: 2,
        feature_count: 2,
    })
    .unwrap();

    // Train: arm 0 is optimal when context[0] > context[1].
    for _ in 0..500 {
        let c0: f64 = rand::Rng::gen_range(&mut rng, 0.0..1.0);
        let c1: f64 = rand::Rng::gen_range(&mut rng, 0.0..1.0);
        let context = [c0, c1];
        let optimal = if c0 > c1 { 0 } else { 1 };

        let arm = bandit.select(&context, &mut rng).unwrap();
        let reward = if arm == optimal { 1.0 } else { 0.1 };
        bandit.update(arm, &context, reward).unwrap();
    }

    // Evaluate: LinUCB should select the context-optimal arm most of the time.
    let mut correct = 0usize;
    let eval_steps = 200;
    for _ in 0..eval_steps {
        let c0: f64 = rand::Rng::gen_range(&mut rng, 0.0..1.0);
        let c1: f64 = rand::Rng::gen_range(&mut rng, 0.0..1.0);
        let context = [c0, c1];
        let optimal = if c0 > c1 { 0 } else { 1 };
        let arm = bandit.select(&context, &mut rng).unwrap();
        if arm == optimal {
            correct += 1;
        }
    }
    let accuracy = correct as f64 / eval_steps as f64;
    assert!(
        accuracy > 0.85,
        "LinUCB should achieve >85% accuracy, got {:.2}%",
        accuracy * 100.0
    );
}

#[test]
fn epsilon_greedy_reset_clears_state() {
    let mut bandit = EpsilonGreedy::new(3, EpsilonGreedyConfig::default()).unwrap();
    bandit.update(0, 1.0).unwrap();
    bandit.update(1, 0.5).unwrap();
    assert_eq!(bandit.samples_seen(), 2);

    bandit.reset();
    assert_eq!(bandit.samples_seen(), 0);
    let stats = bandit.arm_stats(0).unwrap();
    assert_eq!(stats.pulls, 0);
}

#[test]
fn ucb1_reset_clears_state() {
    let mut bandit = Ucb1::new(3, Ucb1Config::default()).unwrap();
    bandit.update(0, 1.0).unwrap();
    bandit.update(1, 0.5).unwrap();
    assert_eq!(bandit.samples_seen(), 2);

    bandit.reset();
    assert_eq!(bandit.samples_seen(), 0);
    let stats = bandit.arm_stats(0).unwrap();
    assert_eq!(stats.pulls, 0);
}

#[test]
fn thompson_reset_clears_state() {
    let mut bandit = ThompsonSampling::new(3, ThompsonConfig::default()).unwrap();
    bandit.update(0, 1.0).unwrap();
    bandit.update(1, 0.5).unwrap();
    assert_eq!(bandit.samples_seen(), 2);

    bandit.reset();
    assert_eq!(bandit.samples_seen(), 0);
    let stats = bandit.arm_stats(0).unwrap();
    assert_eq!(stats.pulls, 0);
}

#[test]
fn linucb_reset_clears_state() {
    let mut bandit = LinUcb::new(LinUcbConfig {
        alpha: 1.0,
        arm_count: 2,
        feature_count: 2,
    })
    .unwrap();
    bandit.update(0, &[1.0, 0.5], 1.0).unwrap();
    bandit.update(1, &[0.3, 0.7], 0.5).unwrap();
    assert_eq!(bandit.samples_seen(), 2);

    bandit.reset();
    assert_eq!(bandit.samples_seen(), 0);
    // A should be back to identity.
    let a = bandit.a_matrix(0).unwrap();
    assert!((a[0][0] - 1.0).abs() < 1e-12);
    assert!((a[0][1] - 0.0).abs() < 1e-12);
}

#[test]
fn epsilon_greedy_rejects_invalid_inputs() {
    let mut bandit = EpsilonGreedy::new(3, EpsilonGreedyConfig::default()).unwrap();
    // Invalid arm.
    assert!(matches!(
        bandit.update(5, 1.0),
        Err(RillError::InvalidArm { .. })
    ));
    // Non-finite reward.
    assert!(matches!(
        bandit.update(0, f64::NAN),
        Err(RillError::InvalidReward(_))
    ));
    // Invalid arm_stats.
    assert!(bandit.arm_stats(10).is_err());
}

#[test]
fn ucb1_rejects_invalid_inputs() {
    let mut bandit = Ucb1::new(3, Ucb1Config::default()).unwrap();
    assert!(matches!(
        bandit.update(5, 1.0),
        Err(RillError::InvalidArm { .. })
    ));
    assert!(matches!(
        bandit.update(0, f64::INFINITY),
        Err(RillError::InvalidReward(_))
    ));
}

#[test]
fn thompson_rejects_invalid_inputs() {
    let mut bandit = ThompsonSampling::new(3, ThompsonConfig::default()).unwrap();
    assert!(matches!(
        bandit.update(5, 0.5),
        Err(RillError::InvalidArm { .. })
    ));
    // ThompsonSampling requires reward in [0, 1].
    assert!(matches!(
        bandit.update(0, 2.0),
        Err(RillError::InvalidReward(_))
    ));
    assert!(matches!(
        bandit.update(0, -0.5),
        Err(RillError::InvalidReward(_))
    ));
}

#[test]
fn linucb_rejects_invalid_inputs() {
    let mut bandit = LinUcb::new(LinUcbConfig {
        alpha: 1.0,
        arm_count: 2,
        feature_count: 2,
    })
    .unwrap();
    // Invalid arm.
    assert!(matches!(
        bandit.update(5, &[1.0, 0.5], 1.0),
        Err(RillError::InvalidArm { .. })
    ));
    // Wrong context length.
    assert!(matches!(
        bandit.update(0, &[1.0], 1.0),
        Err(RillError::DimensionMismatch { .. })
    ));
    // Non-finite reward.
    assert!(matches!(
        bandit.update(0, &[1.0, 0.5], f64::NAN),
        Err(RillError::InvalidReward(_))
    ));
    // select with wrong context length.
    let mut rng = ChaCha8Rng::seed_from_u64(0);
    assert!(bandit.select(&[1.0], &mut rng).is_err());
}

#[test]
fn bandit_select_returns_valid_arm() {
    let mut rng = ChaCha8Rng::seed_from_u64(0);

    let eg = EpsilonGreedy::new(3, EpsilonGreedyConfig::default()).unwrap();
    assert!(eg.select(&mut rng).unwrap() < 3);

    let ucb = Ucb1::new(3, Ucb1Config::default()).unwrap();
    assert!(ucb.select(&mut rng).unwrap() < 3);

    let ts = ThompsonSampling::new(3, ThompsonConfig::default()).unwrap();
    assert!(ts.select(&mut rng).unwrap() < 3);
}

#[test]
fn linucb_select_returns_valid_arm() {
    let mut rng = ChaCha8Rng::seed_from_u64(0);
    let bandit = LinUcb::new(LinUcbConfig {
        alpha: 1.0,
        arm_count: 3,
        feature_count: 2,
    })
    .unwrap();
    let context = [0.5, 0.8];
    let arm = bandit.select(&context, &mut rng).unwrap();
    assert!(arm < 3);
}